Upload
lyliem
View
216
Download
0
Embed Size (px)
Citation preview
Employees’ Retirement Choices,
Perceptions and Understanding:
A Review of Selected Survey and Empirical
Behavioral Decision-Making Research
Sponsored by the
SOA Pension Section
Prepared by
Jodi DiCenzo
Behavioral Research Associates, LLC
March 2014
© 2014 Society of Actuaries, All Rights Reserved
The opinions expressed and conclusions reached by the author are her own and do not represent any official position or opinion
of the Society of Actuaries or its members. The Society of Actuaries makes no representation or warranty to the accuracy of the
information.
© 2014 Society of Actuaries, All Rights Reserved Behavioral Research Associates 2 | P a g e
Table of Contents
Summary ______________________________________________________________ 4
Early Retirement Benefits and the Trend to Defined Contribution Plans ____________ 7
Employees’ Perceptions of Employer-Provided Retirement Benefits _______________ 8
Retirement Benefits and Job Choice ________________________________________ 10
The Pension-Pay Tradeoff _______________________________________________ 11
How Well Do Employees Understand Their Retirement Benefits? ________________ 13
Employee Plan-Type Preferences __________________________________________ 16
Retirement Decisions During the Working Years _____________________________ 22
The Participation Decision _____________________________________________ 23
The Contribution Decision _____________________________________________ 27
Factors Affecting Participation and Contribution Levels _____________________ 28
Enrollment-Related Features _________________________________________ 28
Automatic Deferral Increase Programs _________________________________ 35
Employer Matching Contributions ____________________________________ 36
Investment-Related Features _________________________________________ 40
Loan Provisions ___________________________________________________ 41
Other Retirement Benefits ___________________________________________ 42
Social Norms _____________________________________________________ 42
Other Observed Heuristics and Biases Affecting Saving Decisions ___________ 43
Retirement Plan Investment Decisions ___________________________________ 44
Effects of the Investment Option Menu on Participant Choice _______________ 48
Effects of Investment Performance on Investment Choice __________________ 51
Status Quo Bias and Default Acceptance _______________________________ 53
Employer Stock ___________________________________________________ 54
Leakage ___________________________________________________________ 57
The Effects of Workplace Financial Education _______________________________ 61
Financial Decisions at Retirement _________________________________________ 65
The Annuity Puzzle ____________________________________________________ 74
Conclusion ___________________________________________________________ 77
© 2014 Society of Actuaries, All Rights Reserved Behavioral Research Associates 3 | P a g e
Acknowledgements
The author would like to acknowledge the Project Oversight Group (POG) who provided advice
during the course of this research project.
The members of the POG were:
Sandy Mackenzie (Chair)
Vickie Bajtelsmit
Dick Davies
John Deinum
John Gist
Howard Iams
Cindy Levering
Betty Meredith
Anna Rappaport
Joe Tomlinson
John Turner
Jack VanDerhei
Wendy Weiss
© 2014 Society of Actuaries, All Rights Reserved Behavioral Research Associates 4 | P a g e
Summary
The evolution of workplace retirement benefits in America has greatly altered the roles of the
employer and employee. A system in which paternalistic employers shoulder the burden and
risks of funding retirement for long-tenured workers has largely given way to one in which
employees are fully responsible for funding retirement themselves.
The increased individual involvement in planning and providing for retirement within the context
of a workplace retirement plan has been fertile ground for behavioral economists and other
researchers. Their work has provided a greater understanding of retirement-related decision
making that often stands in stark contrast to the decisions predicted by traditional economists
assuming fully rational agents. These researchers have also offered behaviorally based
prescriptions with the goal of improving workers’ retirement outcomes. The results of their
selected work are discussed here.
While much can be gained from research outside the U.S. retirement domain, the review here is
limited to U.S.-based retirement-specific work. Where it may otherwise not be obvious, the
nature and context of the research is provided in an attempt to minimize inappropriate
extrapolation of research results.
Research context and methodology are critical to appropriate interpretation of results.
Correspondingly, we do not hypothesize about possible underlying psychological principles that
may help to explain research results unless the original research author has done so. The body of
research available has resulted in greater discussion of private-sector defined contribution
retirement plans.
Many important questions like which type of plan is “better,” how retirement-ready workers are
expected to be, or how pensions affect retirement timing and financial decisions in retirement are
not addressed. A complete snapshot view of employees’ retirement choices, perceptions and
understanding is provided. Significant findings are reported below in the order in which they
appear in the main text.
Employees value, and may feel entitled to, retirement benefits.
Experimental evidence suggests that the provision of retirement benefits plays a role in
job choice, but perhaps not as strongly as Ippolito’s sorting theory suggests.
Although the concept of pension-pay trade-off may be less relevant in today’s
environment where (less generous) defined contribution plans are more prevalent, recent
research favors a conclusion that trade-offs are unlikely and imperfect to the extent they
do exist. Older research shows mixed results.
Alarmingly, a significant portion of workers do not seem to understand their retirement
benefits very well (even their plan type), and there has been little improvement over time.
This lack of knowledge does not bode well for optimal retirement plan decision making
nor for the interpretation of research based solely on survey results.
© 2014 Society of Actuaries, All Rights Reserved Behavioral Research Associates 5 | P a g e
Analyses of workers’ plan-type preferences (mostly in public-sector plans) show the
importance of context. While employees tend to make passive choices, in one study, only
a minority do. Suboptimal choices are impacted by inertia, framing, peer effects and
market performance. Younger, higher-income and white employees show a greater
preference for defined contribution plans.
Less than 40 percent of all workers participate in any kind of workplace retirement plan.
This is partially attributable to a lack of access; less than 50 percent of the American
workforce has access to a plan. Older, married, more educated, higher-income and white
employees are more likely to participate.
Less than 20 percent of private-industry workers participate in a defined benefit plan;
nearly 80 percent of public-sector employees do. Approximately 40 and 15 percent of
private and public workers participate in a defined contribution plan, respectively.
Conditional on 401(k) and other salary-deferral plan participation, the approximate
median contribution rate is 5 percent.
Age, income, tenure, education and home ownership are positively associated with
participation in and contributions to a defined contribution plan.
Decision-making context has a dramatic impact on participation and contribution
decisions. For example, reframing or simplifying the enrollment process has improved
participation rates by 10 to more than 40 percentage points in the first year of eligibility.
Prohibiting inertia by requiring a participation decision (either positive or negative) has
increased enrollment rates by nearly 30 percent.
Other plan features specifically designed to thwart suboptimal decision-making
tendencies can also be effective at increasing savings rates. One such feature that
automatically increases savings rates periodically was responsible for nearly quadrupling
average savings rates at one firm.
Incentives are another way to potentially impact participation and contribution behavior.
Within defined contribution plans, the incentive is often in the form of an employer
“matching” contribution equal to some portion of an employee’s contribution, typically
limited to a specified percentage of the employee’s salary. Researchers find positive
participation effects from employer matching contributions. The effect on contribution
rates is less clear. The match threshold is influential, but higher match rates are often
associated with lower contribution rates.
The ability to choose one’s retirement plan investments is associated with higher levels of
plan participation, but the extent of choice also has an effect. Researchers have found that
the total number of options and the number of equity options are associated with lower
participation levels. The presence of company stock predicts higher participation levels,
perhaps offering participants at least one investment they recognize.
© 2014 Society of Actuaries, All Rights Reserved Behavioral Research Associates 6 | P a g e
Researchers have demonstrated that social norms, anchors, past market performance and
priming can also have an impact on savings rates.
In aggregate, asset allocation of 401(k) accounts approximates that of the typical defined
benefit plan: 60 percent invested in equities and 40 percent invested in fixed income.
Generally, equity allocations are positively related to income and negatively related to
age.
Participants’ investment choices are strongly influenced by the choice set offered and
historical investment performance, often in irrational ways.
Across all age groups, a nontrivial percentage of participants allocate more than 20
percent of their retirement accounts to company stock, causing these participants to be
poorly diversified (and subject to unnecessary risk). Researchers suggest a number of
factors explain this phenomenon: perceived implicit advice from the employer, a
familiarity bias, mental accounting, performance chasing and financial illiteracy.
Leakage in the form of preretirement withdrawals and unpaid loans can jeopardize
workers’ retirement security. A cash-out from one’s first defined contribution plan can
result in a 67 percent reduction in his retirement income.
It is estimated that over 50 percent of lump-sum distributions by under age 60
participants are partially (41.4 percent) or wholly (13.4 percent) spent. Younger,
nonwhite, unmarried, lower-income and less-educated participants are more likely to
spend all or a part of their distributions. Smaller distributions are also more likely to be
spent.
Plan loans are taken by approximately 20 percent of participants who have access to
them. Loans result in a relatively minor impact on retirement income, unless the loan is
not repaid, which most often occurs when a participant terminates with an outstanding
loan. In 80 percent of these cases, the participant fails to repay the loan.
Workers’ low level of financial literacy is undisputed. Positive effects of workplace
financial education are reported, but most of the research analyzing actual behavioral
change rather than self-reported surveys suggests the effects are statistically insignificant
or small in absolute terms.
Research that covers retiring participants’ payout choices is limited, and wide variation in
research results suggests contextual effects. Between 12 and 96 percent of retirees choose
lump-sum distributions in the studies presented herein.
Retirement payout choices appear to suffer from some of the same decision-making
shortcomings observed during the accumulation phase. Of particular interest to
researchers is the “annuity puzzle,” the unexplained low levels of interest in annuities.
Concluding that rational explanations fail to fully explain retirees’ decisions, researchers
© 2014 Society of Actuaries, All Rights Reserved Behavioral Research Associates 7 | P a g e
have turned to behavioral explanations, suggesting that loss aversion, an endowment
effect, mental accounting and framing may negatively impact retirees’ annuity purchase
behavior.
Early Retirement Benefits and the Trend to Defined Contribution Plans1
Employer-provided retirement benefits have a long history in the United States. The Employee
Benefit Research Institute (EBRI) documents The New York City Teachers Retirement Plan,
introduced in 1869, as the first public-sector retirement-income plan (1998). The American
Express Company commenced one of the first private-sector plans in 1875 (Greenough & King,
1976). Both of these plans were defined benefit plans, but the American Express plan differed
significantly from today’s traditional defined benefit plans. Specifically, the general manager had
to approve an employee’s retirement, and benefits were paid only to disabled, elderly workers.
The benefit for an eligible plan member was set at 50 percent of average earnings over the last 10
years of work, with a cap of $500 (Latimer, 1932).
While some of the first workplace plans may have provided defined benefits funded by the
employer, there is also early evidence of workers being responsible for funding their own
retirement, as evidenced by a carriage shop’s 1880 posting of employee rules:
Working hours shall be from 7 a.m. to 9 p.m. every day except the
Sabbath. … After an employee has been with this firm for five
years he shall receive an added payment of five cents per day,
provided the firm has prospered in a manner to make it possible.
… It is the bounden duty of each employee to put away at least 10
percent of his monthly wages for his declining years so he will not
become a burden upon his betters. (Milkovich & Newman, 2002)
Still, in contrast to many other industrializing nations where individuals were provided protective
benefits by the government, many of America’s employers began to shoulder the welfare of their
employees by providing health and retirement benefits, among others.
Shaped by economic, regulatory, cultural and demographic forces, the nature and form of
employment-based retirement benefits in America have evolved since their inception.2 Most
notably, private employers have increasingly moved away from providing retirement benefits
that promise a steady stream of income in retirement (a defined benefit plan) to offering
employees defined contribution plans such as profit sharing, 401(k) and money purchase plans,
or a combination of both.3,4,5 From 1975 through 2009, the number of private-sector defined
1 Here, and throughout this paper, “retirement benefits” refers to voluntary retirement benefits and does not include Social
Security. 2 For overviews of the evolution of workplace retirement benefits, see EBRI Databook on Employee Benefits, Chapter 1 and
Appendix E, and EBRI Facts, U.S. Retirement Income System, December, 1998. 3 For a description of the different types of retirement plans, see http://www.dol.gov/dol/topic/retirement/typesofplans.htm (U.S.
Department of Labor, 2013). 4 These same trends are not replicated in the public plan domain. Bovbjerg (2008) finds that as of 2007, virtually every state
offered a defined benefit plan as its primary benefit. Only two states and the District of Columbia offered a defined contribution
plan as their primary plans. In addition, 2012 National Compensation Survey statistics show that 83 percent of state and local
© 2014 Society of Actuaries, All Rights Reserved Behavioral Research Associates 8 | P a g e
benefit plans declined by more than half—from approximately 103,000 to 47,000 plans while the
number of defined contribution plans more than tripled from approximately 208,000 plans to
nearly 660,000 plans. During this same period, the percentage of active private-sector employees
(with a retirement plan available to them) eligible to participate in defined benefit plans declined
from 71 percent to 20 percent (U.S. Department of Labor, 2011). EBRI (2013) estimates an 89
percent (55 percentage point) decline—from 62 percent in 1975 to 7 percent in 2011—in the
number of employees participating in a defined benefit plan offered as the exclusive retirement
benefit plan.6 Figure 1 below shows the participation trend by plan type.
Figure 1. Retirement plan trends: participation by plan type, private-sector, active-worker
participants, 1979–2011
Employee Benefit Research Institute (2013)
Original source: U.S. Department of Labor Form 5500 Summaries 1979–98. Pension Benefit Guaranty Corp.
Current Population Survey 1999–2011, EBRI estimates 1999–2010.
Employees’ Perceptions of Employer-Provided Retirement Benefits
An important empirical question is the extent to which employees value retirement benefits.
According to three recent surveys, employees do appear to qualitatively appreciate the retirement
benefits offered by their employers. While the surveys below seek preferences of private-sector
employees, Fredericksen and Soden (1998) found similar preferences for both private- and
public-sector employees.7
In the Principal Financial Well-Being Index survey conducted in the fourth quarter of 2011, 69
percent of respondents rated their defined contribution retirement plan benefit as one of the more
government employees have access to a defined benefit plan and 31 percent have access to a defined contribution plan (U.S.
Department of Labor, 2012). 5 See chapter 7 in Pension Plans and Employee Performance (Ippolito, 1997) and chapter 8 in The Choice of Pension Plans in a
Changing Regulatory Environment (Clark & McDermed, 1990) for discussions of potential explanations for the shifts. 6 Also U.S. Department of Labor, Form 5500 Filings, Pension Benefit Guaranty Corporation, Bureau of the Census, and Current
Population Survey, all as cited by EBRI (2011). 7 The research sample was restricted to employees in El Paso, Texas.
© 2014 Society of Actuaries, All Rights Reserved Behavioral Research Associates 9 | P a g e
important benefits provided by their employers (Principal Financial Group, 2011).8 The second-
place ranking to health care has been consistent but declining (from 75 percent of employees in
2006) over the last five years. Defined benefit pension plans were ranked as an important benefit
by 48 percent of respondents, a percentage that was also lower than in prior years when the
percentage ranged from 57 to 51 percent.9 Interestingly, the top “wished for” benefit was a
defined benefit plan, which was selected by nearly 25 percent of respondents.
The 2011 Employee Job Satisfaction and Engagement (Society for Human Resource
Management (SHRM)) showed similar results: Defined contribution plans and defined benefit
plans ranked third and fourth, respectively, as the most important benefits offered by
respondents’ employers.10 (Health care benefits and paid time off were ranked higher.) The
survey results also showed that African-American employees (as compared to Caucasian
employees) and baby boomers (as compared to Gen Xers) placed greater importance on
retirement benefits. In addition, defined benefit plans were found to be more important to large
company workers. (This is likely because defined benefit plans are more prevalent in larger
companies.)
Finally, 66 percent of respondents to a 201011 telephone survey jointly conducted by EBRI and
the Financial Services Roundtable (FSR) assigned a top rating to retirement savings accounts for
their importance in providing financial protection. Importance ratings were higher among older
and higher-income workers—rising to 80 percent in some segments. The highest importance
rating was also assigned to pensions by 55 percent of respondents (EBRI & FSR, 2010).
Despite being viewed as valued benefits, some research suggests that employees view benefits as
entitlements, or part and parcel of their employment, regardless of their or the company’s
performance (Williams, 1993; Hart & Carraher, 1995; Sinclair, Hannigan & Tetrick, 1995).
Weathington and Tetrick (2000) further explore this concept to separately determine the degree
to which workers believe employees are entitled to each type of benefit.12 They indeed find a
sense of entitlement that extends to retirement plan benefits. Nearly 77 percent of subjects rated
their agreement with the statement that a workplace retirement plan is an entitlement a 5 or better
on a scale from 1 to 7 (strongly agree).13,14
8 The online survey was conducted in October 2011 within the United States by Harris Interactive. Respondents—1,121
employees and 533 retirees—were selected for Harris’ online panel. The data have been weighted to reflect the composition of
the entire population of retirees and adult employees working for small to mid-sized U.S. businesses. “Because the sample is
based on those who agreed to be invited to participate in the Harris Interactive online research panel, no estimates of theoretical
sampling error can be calculated” (Principal Financial Group, 2011). 9 Note that these percentages represent the percentages of total respondents who assigned an 8, 9 or 10 importance rating to the
benefit on a 1- to 10-point scale. 10 The survey is from a randomly selected sample from a third-party Internet panel based on the American Community Study. A
response rate of 83 percent yielded 600 participants who were employed full or part time. The survey was conducted in 2011. 11 The survey was conducted in September 2010 by the Opinion Research Corporation. A sample of 1,007 adults age 18 and
older, living in private households in the continental United States were interviewed. 12 Subjects included 216 employed undergraduate students attending a large southern university. Ages ranged from 19 to 73; 72.7
percent were women. A majority worked on a full-time basis (61.1 percent). Approximately 5 percent worked both a full- and a
part-time job. 13 Of the benefits tested, retirement plan benefits were associated with the lowest sense of entitlement, behind medical insurance,
paid holidays, paid vacation, paid sick leave and family leave. 14 The researchers offer no explanation for entitlement perceptions but find that the higher degree of entitlement perceptions, the
more closely positive the association between benefit satisfaction and affective commitment and organizational commitment.
© 2014 Society of Actuaries, All Rights Reserved Behavioral Research Associates 10 | P a g e
Retirement Benefits and Job Choice
A pat explanation for the provision of employer benefits is “to attract and retain quality
employees.” Ippolito (1997) extends this with his notion of sorting, which suggests employers
use pensions to sort prospective employees, implying that pensions matter in job choice.
Underlying this concept is the assumption that prospective workers who are attracted to pensions
have qualities associated with good performance. His sorting theory not only suggests that
pensions can be used to sort workers generally, but that the type of pension offered can sort
workers. Ippolito posits that high discounters (workers with high discount rates) will not be
attracted to jobs with defined benefit pension plans, but will more likely prefer jobs with defined
contribution plans. Ippolito cites Curme and Even (1995) who show a connection between
spending propensities, credit constraints and defined benefit plan coverage as empirical evidence
of his theory.
Although there is some evidence that supports Ippolito’s sorting theory (see section on
participation decisions), it is not found in early survey work in this area. Surveys tended to only
address benefits overall (not specific types of benefits) as a basis for job selection and found that
benefits were unimportant. This may have been a result of the nature of the research as well as
the subject pool as initial studies involved surveying graduate students. Against opportunity for
advancement, pay, location, responsibilities and prestige, benefits were ranked last (Huseman,
Hatfield, & Driver, 1975; Huseman, Hatfield & Robinson, 1978). However, Barber and Roehling
(1993) find that subjects gave most attention to location, salary and benefits in their reviews of
job opportunities for the purpose of hypothetically deciding to interview and devoted even more
attention to benefits when they were generous.
More recently, Tetrick, Weathington, Da Silva, and Hutcheson (2010) studied the impact of
several aspects of a total compensation package (e.g., salary, vacation time, health insurance cost
and type of retirement plan) on hypothetical job preferences, albeit with an unrepresentative
subject pool.15 Subjects were provided 32 job descriptions that varied starting salary, vacation,
cost of health insurance and type of retirement plan (if any). The conditions included four
variations of retirement plan offering: none, a 401(k), a defined benefit and a “company stock”
plan. As one would expect, subjects were more likely to apply for and accept jobs that are higher
paying, offer more vacation, and where the employees’ share of health insurance costs are lower.
They were also more likely to apply and accept the job if a retirement plan were offered, but it
made no difference what type of retirement plan it was.
Tested predictors of the importance of the various compensation factors included self-reported
age, marital status, previous benefit history, risk aversion, achievement orientation, attitudes
toward earnings and consideration of future consequences. Only retirement-related results are
presented here, and the comparison group for all is “no retirement plan.”
Married subjects were more likely to value a defined benefit plan.
15 Sample is 76 university students, most (69) who were undergraduate students. Ages ranged from 18 to 50, with a mean age of
25. Approximately 62 percent of subjects were women. Six students were unemployed, 24 worked full time, 39 worked on a part-
time basis, and two held both a full- and a part-time job. Average prior year earnings ranged from $15,000 to $19,999.
© 2014 Society of Actuaries, All Rights Reserved Behavioral Research Associates 11 | P a g e
Benefit history was positively related to the preference for defined benefit and company-
stock-funded plans.
Subjects with a greater propensity for risk preferred 401(k) and company-stock-funded
plans.
Achievement orientation positively correlated with the relative importance of a company-
stock-funded plan.
Attitudes toward earnings pointed to a greater preference for defined benefit and 401(k)
plans.16
The extent to which future consequences are considered was predictive of a preference
for a 401(k) plan.
In commenting on the surprise finding of no age-related preferences, they note this may relate to
the fact that the oldest member of the subject pool was 50.17
The Pension-Pay Trade-Off
Next we turn attention to empirical evidence of employees’ trade-off between current and
deferred compensation (retirement benefits). In other words, how much current pay are
employees willing to give up in exchange for future retirement income? Labor economists have
theorized that in a competitive labor market, compensation among similar workers will be
equalized.18 Similar workers in similar jobs will receive similar compensation, albeit in different
forms due to heterogeneous preferences for cash compensation and benefits, therefore predicting
relatively straightforward, dollar-for-dollar trade-offs between benefits and pay (Brown 1980).
Stated differently, rational workers only forgo current pay to the extent that they are equivalently
compensated with other benefits. Compensation practices are also often driven by this logic as
managers work from a total compensation package, assuming rational employees accepting equal
trade-offs between pay and fringe benefits.
However, empirical evidence proving this rational theory is scant—in some cases owing to the
difficulty of obtaining complete and reliable data that include all relevant variables (Smith &
Ehrenberg, 1983; Gustman & Mitchell, 1990). In fact, many early researchers in this area
specifically mention limitations of their research, and interested readers are encouraged to review
specific sources for additional details. Given researchers’ concern with data reliability and
completeness, it should come as no surprise that the results of early research are mixed, as
evidenced in Table 1 below, which reports older work included in Gunderson, Hyatt and Pesando
(1992).
16 Attitudes toward earnings were based on responses to nine questions using a five-point Likert-type response scale to measure
the value placed on earning money. 17 They also refer to Miceli and Lane (1991), who suggest that the preference for protective benefits “may be best predicted by
the joint effect of age and family responsibilities.” As retirement plans seek to “protect” income, they may be considered within
the category of protective benefits. 18 This concept was introduced by Adam Smith ([1776] 1937) who posited “The whole of the advantages and disadvantages of
the different employments of labor and stock must, in the same neighborhood, be either perfectly equal or continually tending
toward equality.”
© 2014 Society of Actuaries, All Rights Reserved Behavioral Research Associates 12 | P a g e
Much of the pension-pay trade-off research is dated, and some researchers believe the question
of whether a pension-pay trade-off exists is less relevant in today’s environment of (generally
less generous) 401(k) plans. In fact, more recent research favors a conclusion that a trade-off
does not exist. Only Gerakos (2010) finds an imperfect trade-off of 48 cents of pay for every
pension benefit dollar. 19 Cadman and Vincent (2011) infer that (nonqualified) defined benefit
pension compensation is in addition to and not a substitution for other forms of compensation. In
hypothetical experimental research, Tetrick et al. (2010) find an additional $10,000 per year does
not offset the perceived “value” of an additional three weeks of vacation, having to pay the entire
cost of one’s health insurance (at the company’s rates) or assuming full responsibility of funding
one’s retirement. Weathington (2008) further explores the relationship between income and
willingness to trade retirement benefits for cash and finds no relationship.
Table 1. Studies of wage-pension benefit trade-offs
Study Data Pension Variables Results
Ehrenberg
(1980)
Two data sets on
municipal police,
firefighters, and sanitation
workers in the years 1973-
75
Ratio of pension benefits to
earnings at time of
retirement; employee’s
contributions; measures of
likely underfunding; other
characteristics of pension
plans
Some evidence that
increased employee
contributions and
underfunding lead to
higher wages; limited
evidence of wage-pension
benefit trade-off (police in
one data set); mixed results
for other characteristics
Schiller and Weiss
(1980)
1969 pension file linked
with Social Security
earnings for men
Ratio of pension cost to
wages; other pension plan
characteristics
Trade-off for 3 of 5 age
groups, but significant only
for 45-54 age; mixed
results for other
characteristics
Smith (1981) Government employees in
86 cities in Pennsylvania
in 1976
Pension benefit accrual and
measure of pension
underfunding
Significant trade-offs,
usually insignificant
Smith and Ehrenberg
(1983)
193 firms with wage and
pension differences across
jobs of different Hay point
job evaluations
Differences in pension
value across jobs in
different Hay Scores
No significant trade-offs
Bulow and Landsman
(1985)
1982 data on 993 faculty
at Stanford
Probability of signing up
for pension plan
Weak trade-off, usually
insignificant
Clark and McDermed
(1986)
1971/73/75 Retirement
History Survey, men
Working past age of
normal retirement and
hence experiencing
negative pension accruals
Trade-off in the sense that
a significant compensating
wage premium is
associated with expected
pension loss from delayed
retirement
Mitchell and Pozzebon
(1986)
1,696 employees, 666 with
pension plans, from 1983
Survey of Consumer
Finance
Coverage by pension plan;
pension contributions; and
other pension plan
characteristics
No trade-off; more often, a
wrong-signed relationship
19 The researcher’s sample included 442 chief executive officers of S&P 500 companies. Defined contribution plan benefits have
been excluded from his estimates of pension benefits.
© 2014 Society of Actuaries, All Rights Reserved Behavioral Research Associates 13 | P a g e
Study Data Pension Variables Results
Gustman and Steinmeier
(1987)
558 full-time private
sector men from 1983
Survey of Consumer
Finance
Coverage by pension plan Significant positive
relationship
Moore (1987) 4,500 employees from 5
firms
Pension cost to employer Significant negative trade-
off under 2SLS to account
for the fact that pensions
are a positive function of
wages in earnings-based
plans; significant positive
relationship under OLS
Dorsey (1989) 1,973 full-time private
sector employees from
1983 Survey of Consumer
Finance
Coverage by pension plan,
simultaneously determined
Significant positive
relationship in both OLS
and 2SLS, the latter to
account for the possibility
that pension coverage is a
function of wages
Montgomery, Shaw, and
Benedict (1990)
529 employees with
defined benefit pension
plans, from 1983 Survey
of Consumer Finance
Pension benefit accrual as
% of wages
Significant trade-off, but it
becomes insignificant
when 2SLS is used to
account for simultaneity
Even and Macpherson
(1990)
6,317 employees from
1983 Survey of Consumer
Finance
Coverage by pension plan Significant positive
relationship
Gunderson (1992) 98 matches pension plans
and collective agreements,
Ontario
Actuarial calculation of
employer’s expected
pension cost, and; pension
plan characteristics
affecting that cost
Significant trade-off,
especially for flat benefit
rate, but not for early and
postponed retirement
provisions; trade-off only
when pension variable
specified as replacement
rates, not amounts
Note. From “Wage-pension trade-offs in collective agreements,” by Gunderson, Morley, D. Hyatt, and J.E. Pesando,
1992, Industrial and Labor Relations Review, p. 148. ©1992 Cornell University. Reprinted with permission.
How Well Do Employees Understand Their Retirement Benefits?
There are at least three reasons why employees’ understanding of their retirement benefits is
important. As Gustman, Mitchell and Steinmeier (1994) note, for pensions to have the desired
effects on employee behavior, workers must understand them, including “the risks they face, and
value the insurance the pension provides.” Second, to the extent that large groups of employees
lack a basic understanding of their retirement benefits, it is naive at the outset to assume them
rational agents able to optimally plan for retirement. Third, many of the datasets used in
retirement-related research rely, at least in part, on accurate respondents.
Unfortunately, significant empirical evidence suggests that retirement benefits are quite poorly
understood, even among workers who are nearing retirement and dependent upon them.
Additionally, little improvement has occurred over the 30 years since Mitchell (1988) and
© 2014 Society of Actuaries, All Rights Reserved Behavioral Research Associates 14 | P a g e
Gustman and Steinmeier (1989) began analyzing 1983 Survey of Consumer Finances (SCF) data
by comparing respondent answers to information collected from employers, as shown in Table 2.
Table 2. Summary of pension knowledge research
20 Starr and Sunden were employed by the Federal Reserve Board of Governors. The Federal Reserve Board conducts the Survey
of Consumer Finances. 21 Starr and Sunden (1999) admit that in their study, the assumptions employed in matching 1989 SCF data to employer records
give the respondent the benefit of the doubt, which may cause some upward bias in respondent accuracy. 22 One might expect that as defined contribution plans gained in popularity, individuals would become more accurate in their
responses. However, Dushi and Honig (2008) find that 2004 Health and Retirement Study (HRS) respondents were no more
accurate in correctly reporting whether they contribute to a defined contribution plan than the original 1992 cohort was. These
authors note that both cohorts had a tendency to overstate their contributions. Over 20 percent of both cohorts reported they
contributed to a defined contribution plan but did not according to their W-2 reports.
Author and Data Key Findings
Gustman and Steinmeier (1989)
1983 SCF data
Conditional on employer reporting of plan type, 63
percent of employees covered by a defined benefit plan
correctly said so.
For those covered by defined contribution plans, the
corresponding percentage was 37 percent.
Mitchell (1988)
1983 SCF data
Unionized, higher-income, longer tenured and more
educated respondents tend to be better informed.
Starr and Sunden (1999) 20
1989 SCF data
Over three quarters of respondents could correctly
identify their plan type.
Employees covered by a defined contribution plan (DC
workers) were more likely to know their plan type.21
Fewer DC workers knew whether they themselves
contributed than knew whether their employer
contributed.22
Less than a third of DC workers knew whether their plan
had any withdrawal provisions, but 60 percent of them
correctly knew their plans’ loan provisions.
Of those covered by a defined benefit plan, 75 percent
knew the basic contribution provisions of the plan and 80
percent knew their vesting status.
Gustman and Steinmeier (2004)
1992 Health and Retirement Study (HRS) data matched
with Social Security earnings records as well as
employer retirement plan descriptions
Although 77 percent of respondents correctly identify
that they are in a defined benefit-type plan, widespread
misinformation exists.
Those who are most dependent on pensions are better
© 2014 Society of Actuaries, All Rights Reserved Behavioral Research Associates 15 | P a g e
Table 3. Summary of respondent and firm reports over time23
1983 1992 2004
1. DB only: employer data 88 48 25
2. DC only: employer data 8 21 26
3. Both: employer data 3 31 49
4. DK 16 2 3
5. Frequency of DB only Understated by 18
percentage points in
respondent report, after
Understated by 1
percentage point in
respondent report, after
Overstated by 16
percentage points in
respondent report, after
23 Lines 1 through 3 show the frequency of plan type reported in employer data. Line 4 represents the fraction of employees who
do not know their plan type.
informed; 93 percent of those whose pension wealth
represents 60 percent or more of their total wealth can
correctly identify their plan type.
Women are 7 percent less likely to correctly identify
their plan type, as are older respondents. However,
women are more likely to say they don’t know and older
individuals are less likely to do so.
More than 40 percent of respondents do not know what
their pension is worth.
Gustman, Steinmeier and Tabatabai (2010)
1983 Survey of Consumer Finances, 1992 through 2004
Health and Retirement Studies, employer and consulting
firm data
A significant portion (over one-third) of workers do not
know the type of retirement plan in which they
participate.
The lack of plan knowledge has persisted over time (see
Table 3) and extends to workers approaching retirement.
Dushi and Iams (2010)
Comparison of 1998 and 2006 Survey of Census
Bureau’s Survey of Income and Program Participation
(SIPP) data to W-2 wage reports
While 40 percent of 1998 and 39 percent of 2006
respondents said they contributed to a defined
contribution retirement plan, 46 percent actually did so.
The gap between actual (49 percent) and reported (37
percent) participation was even wider in the public
sector.
Clark, Morrill and Allen (2012)
Survey of 1,500 workers nearing retirement from three
large companies (average age of 56.5)
Fifty-six percent did not know their expected pension
income as a percentage of their salary.
Only 36 percent knew the earliest age they could receive
their pension.
Only 16.2 percent knew the reduction if retirement
benefits were taken at an early age.
© 2014 Society of Actuaries, All Rights Reserved Behavioral Research Associates 16 | P a g e
1983 1992 2004
excluding DKs excluding DKs excluding DKs
6. Frequency of DC only Overstated by .4
percentage points in
respondent report, after
excluding DKs
Overstated by 3
percentage points in
respondent data, after
excluding DKs
Overstated by 7
percentage points in
respondent data, after
excluding DKs
7. Frequency of both Overstated by 17
percentage points in
respondent report, after
excluding DKs
Understated by 2
percentage points in
respondent data, after
excluding DKs
Understated by 23
percentage points in
respondent data, after
excluding DKs
8. Share on diagonal:
employer and respondent
agree
60 percentage points on
diagonal (despite high
DK)
49 percentage points on
diagonal
45 percentage points on
diagonal
9. Conditional on firm
reporting DB only,
respondent reporting DC
only
4 15 20
10. Conditional on firm
reporting DB only,
respondent reporting both
17 27 27
11. Conditional on firm
reporting DC only,
respondent reporting DB
only
29 26 14
12. Conditional on firm
reporting DC only,
respondent reporting both
only
10 18 14
13. Conditional on firm
reporting both, respondent
reporting DB only
35 45 48
14. Conditional on firm
reporting both, respondent
reporting DC only
10 18 19
Note. From Pensions in the Health and Retirement Study. by A.L Gustman, T.L. Steinmeier, & N. Tabatabai, 2010,
Cambridge, MA: Harvard University Press. Reprinted with permission.
Much of the research reported above was driven by researchers’ desire to better understand (and
improve) the reliability of the large data sets available for study. What they find about
individuals’ knowledge level of their own retirement benefits is telling. Many workers don’t
know the type of plan they’re in or whether they are contributing to it. This offers an important
backdrop for the exploration of workers’ retirement decision-making journeys covered herein.
Employee Plan-Type Preferences
There are few opportunities to empirically analyze workers’ preferences for one type of
retirement plan over another. Once hired, a relatively small percentage of employees have the
opportunity to choose between a defined benefit plan and a defined contribution plan. When they
© 2014 Society of Actuaries, All Rights Reserved Behavioral Research Associates 17 | P a g e
do, it most often occurs within the public sector, as is evidenced by the extent of available
research reviewed here. All except one of the papers examine employee plan-type preferences
within the public sector. Specifically, most focus on faculty from state-funded university systems
where it is more common for members to be given the option of choosing their pension plan
upon hiring.24
In this section, we report on five academic studies of employees’ actual decision-making.25 Two
of the studies analyze choices of newly hired employees, and three others focus on the decisions
of employees who have been given a one-time opportunity to switch from their existing defined
benefit plan. As can be seen from Table 4 below, Brown and Weisbenner (2007) and Clark,
Ghent and McDermed (2006) study the decisions of new employees, whereas Benartzi and
Thaler (2007), Yang (2005) and Papke (2004) study the decisions of participants given the
opportunity to switch from a traditional defined benefit plan to a defined contribution plan.
Major findings are highlighted below.
The mixed results of these studies suggest the importance of context in decision-making
outcomes. For example, we see significant evidence of passive choices in Brown and
Weisbenner (2007), Benartzi and Thaler (2007), Yang (2005) and Papke (2004), but in
Clark et al. (2006) over 80 percent actively choose their plan type.
Authors suggest that a nontrivial portion of employees make suboptimal choices
regardless of whether the choice was actively or passively made (Brown & Weisbenner,
2007; Benartzi & Thaler, 2007; Yang, 2005). Brown and Weisbenner (2007) find that
higher-income, more educated employees actively choose to participate in a defined
contribution plan even though the authors’ analysis showed the portable defined benefit
plan to be a superior option. Benartzi and Thaler (2007) find that lower-tenured
employees passively accept their continued participation in the defined benefit plan
offered, even though the likelihood of breaking even (as compared to participating in the
newly offered defined contribution plan) is 13 percent.
Benartzi and Thaler (2007) partially attribute suboptimal choices to inertia, citing that
while only 10 percent expected to be defaulted into the defined benefit plan (based on
advanced survey results), 63 percent of employees actually were.
Brown and Weisbenner (2007) and Yang (2005) suggest that the way information was
framed impacted employee choices.
Peer effects are also implicated as contributors to suboptimal choices (Brown &
Weisbenner, 2007; Clark et al., 2006).
24 According to a 2007 American Association of University Professors survey, 97 percent of the public colleges and universities
responding offer faculty the option to choose their pension plan, whereas just 3 percent of private schools do (Conley, 2007). 25 In a 2009 report by consultant Mark Olleman on the decisions by new hires in seven public systems, he finds that between 39
and 90 percent are defaulted into defined benefit plans. Between 13 and 43 percent actively choose their defined benefit option,
and between 3 and 26 percent actively choose their defined contribution option (Olleman, 2009). (This is intentionally excluded
from Table 3 above.)
© 2014 Society of Actuaries, All Rights Reserved Behavioral Research Associates 18 | P a g e
Employees’ choices are also thought to be influenced by favorable recent market
performance, which is associated with a greater preference for defined contribution plans
(Brown & Weisbenner, 2007; Benartzi & Thaler, 2007), but this does not appear to be the
case in Clark et al. (2006).
Analyzing choices by various demographic variables, researchers show that high earners
tend to prefer defined contribution plans, women are more likely than men to take action
(Brown & Weisbenner, 2007; Yang, 2005; Papke, 2004), older employees are more likely
to choose defined benefit plans (Brown & Weisbenner, 2007; Clark et al., 2006; Yang,
2005), and whites are more likely to choose defined contribution plans (Clark et al.,
2006; Yang, 2005).
Survey work related to plan-type preferences is also included in Table 3 and briefly summarized
below.
Table 4. Summary of plan choice research
Author and Study Default
Option Key Findings
Studies Using Actual Behavioral Data
Brown and Weisbenner (2007)
This study covers decisions of 45,000 state university
employees, including campus administrators, university
police, etc., hired between 1999 and 2004. Data are
extracted from the State University Retirement System
(SURS) of Illinois.
Within six months of employment, employees have a one-
time, irrevocable choice between a traditional DB plan, a
portable DB plan or an entirely self-managed DC plan.
Subjects are not covered by Social Security.
Defined
benefit plan
56 percent were defaulted into the
traditional DB plan
18.9 percent chose the portable DB
plan
15.3 percent chose the DC plan
10 percent actively chose the
traditional DB plan
Benartzi and Thaler (2007)
Researchers study the behaviors of (undisclosed) public-
sector employees who, in the latter half of 2002, are given
the opportunity to select either a new defined contribution
plan (with a vesting period of one year) or a hybrid plan as
an alternative to remaining in the defined benefit plan,
which had a six-year vesting schedule. Results presented at
right are as of early February 2003.
Defined
benefit plan
63 percent defaulted into the DB plan
7 percent of employees with less than
two years of tenure switched to the
DC plan
Clark et al. (2006)
This study examines retirement plan choices of 7,035 new
tenure-track employees hired between 1983 and 2001 into
the University of North Carolina school system, across 15
campuses.
Defined
benefit plan
83.8 percent chose the DC plan and
16.2 percent chose the DB plan
Authors do not indicate portion that
made active versus passive decision
© 2014 Society of Actuaries, All Rights Reserved Behavioral Research Associates 19 | P a g e
Author and Study Default
Option Key Findings
Within 30 days after being hired or 30 days after becoming
an instructor, employees must choose between a defined
benefit and a defined contribution plan, and employees who
do not make an active choice are automatically enrolled in
the DB plan. Employees who enrolled in a DC plan could
switch to another DC plan, but the DB or DC choice was
irrevocable.
Employees are covered by Social Security.
Older, female and nonwhites were
more likely to choose the DB plan
Yang (2005)
This paper analyzes the choices made by 3,535 employees
at a nonprofit firm who were given the one-time opportunity
(between March and June 2000) to switch from their current
DB plan into a new DC plan.
No switch to
DC (remain
in DB)
Half of employees switched to DC
The other half remained in DB, but 6
percent made an active choice (to
remain in the plan)
More likely to choose DC: female,
white, higher-income, nonunionized,
familiar with previous DC plan
Papke (2004)
Papke analyzes the choice to remain in a DB plan or transfer
vested balance to DC plan made by 13,170 Michigan state
correctional workers when a one-time, irrevocable offer
between Jan. 2, 1998, through April 30, 1998, to transfer
vested DB balance to a DC plan was made.
In 1997, Michigan closed its DB pension plan to new state
employees. Employees hired in and after 1997 were
defaulted into DC plan.
Employees are covered by Social Security.
No transfer
to DC
(remain in
DB)
Only 1.6 percent of eligible
employees made the switch from DB
to DC
Studies Using Survey Data
DeArmond and Goldhaber (2010)
During spring 2006, 3,080 full-time Washington state
teachers responded to a survey to assess their plan
knowledge and to find out whether they would
hypothetically prefer an additional 10 percent of salary in a
DB or DC plan:
68 percent of respondents reported being in a
hybrid plan
27 percent reported being in a traditional DB plan
4 percent were unsure
Teachers hired after 1996 have been automatically enrolled
in a hybrid DB/DC plan.
Not
applicable
49 percent prefer DC
26 percent prefer DB
26 percent unsure
© 2014 Society of Actuaries, All Rights Reserved Behavioral Research Associates 20 | P a g e
Author and Study Default
Option Key Findings
Respondents are covered by Social Security.
Mathew Greenwald & Associates Inc. (2004)
Reported results are based on mail surveys received in early
2003 from 790 workers primarily between the ages of 25
and 74.
Not
applicable
Sixty-two percent of working DC plan
participants prefer DC plans; only 19
percent of this group prefers DB plans
Fifty-one percent of workers in
defined benefit plans prefer this plan
type; 30 percent of this group prefers
defined contribution plans. The author
notes that this finding is strongly
influenced by government employees.
Dulebohn, Murray and Sun (2000)
This research is based on a field survey of 2,410 employees
who participate in the state-sponsored retirement (defined
benefit) system, from 60 different employers, to identify
determinants of plan choice as well as employee attitudes
and preferences for various plan features.
Not
applicable
41 percent preferred the hypothetical
DB plan
44 percent preferred the hypothetical
hybrid plan
16 percent preferred the hypothetical
DC plan
Clark, Harper and Pitts (1997)
Study is based on 1995 survey data (from 580 North
Carolina State faculty hired between 1971 and 1994)
designed to learn more about employees’ plan choice
decision making, the determinants of their choices and plan
knowledge.
Not
applicable
29 percent didn’t give much thought
to their plan selection
30 percent of respondents said they
hadn’t consulted anyone when making
their decision
Over 40 percent didn’t realize the
DB/DC choice was irrevocable
Brown and Weisbenner (2007) note that a growing portion of the population is defaulted into the
defined benefit plan (from 42 percent in 1999 to 60 percent in 2004). Further, younger, lower-
wage males with lower job attachment are more likely to accept the default. Unfortunately, it is
difficult to know whether these individuals actively decided to do nothing because they knew
they would be defaulted into their preferred plan type (the defined benefit plan). The authors find
that higher earners who are more educated are more likely to actively choose the defined
contribution even though the authors’ analysis suggests the defined contribution plan is
suboptimal to the portable defined benefit plan option. They suggest employees’ choices were
influenced by the way the plan options were framed in the informational materials provided to
employees.26 They also find evidence of the influence of peers and recent market performance.
26 The authors also offer that employees may have been concerned about the political risk associated with state’s pension
underfunding, they may be overconfident about future market returns and their ability to manage their assets, or they may simply
place a high value on being able to manage their own investments (Brown & Weisbenner, 2007).
© 2014 Society of Actuaries, All Rights Reserved Behavioral Research Associates 21 | P a g e
More specifically, an employee choosing a plan type in 2003 (after a significant market decline)
was 11 percentage points less likely to choose the self-managed defined contribution plan than
was a new employee in 1999. Finally, higher earners who are older, married and female working
in community colleges are more likely to actively choose the traditional defined benefit plan
(Brown & Weisbenner, 2007).
Benartzi and Thaler’s (2007) analyses showed that under most scenarios, the defined benefit plan
was a suboptimal choice for younger, lower-tenured employees. In fact, they estimate the
likelihood of breaking even (the defined benefit plan benefit equals the defined contribution plan
benefit) at just 13 percent. The plan actuary estimated that a 31-year-old with one year of tenure
has a 10 percent chance of continuing his career (to age 62) with the same employer. However,
they find that just 7 percent of employees with less than two years of tenure switched to the
defined contribution plan. Further, less than half the participants who planned to switch to the
defined contribution plan actually did.
There are two striking findings in the work of Clark et al. (2006). First is the decline in the
preference for the defined benefit plan and second is the dramatic difference in the preferences of
white and blacks. The authors note that in 1983, the first year studied, 18.2 percent of whites and
60.6 percent of blacks selected the defined benefit plan, whereas in 2001, the comparable
percentages are 9.3 and 32.8. Their regression model estimates that gender, race, age, rank and
classification have significance in predicting plan type preferences of new hires. Women, blacks,
those who are older, those who are not full or tenured professors, and those who do have not
doctorates are more likely to choose the defined benefit plan. The authors suggest that choices of
newly hired blacks may be influenced by peer effects.
The work by Yang (2005) and Papke (2004) studies the decisions by employees who are given
the one-time opportunity to switch or transfer vested defined benefits to a defined contribution
plan. The employees in Yang’s nonprofit firm were much more likely to switch. Each employee
was provided with projected annual benefits in retirement under each plan, assuming varying
levels of tenure. The authors find that people in situations where it was relatively easy for the
defined contribution benefit to “catch up” to the defined benefit were more likely to switch.
Interestingly, they also find that if an employee’s projected benefit at age 65 from the defined
contribution plan exceeded that from the defined benefit plan, he was less likely to switch,
causing the author to believe that perhaps employees did not pay much attention to projected
benefits. In addition, Yang finds the following attributes are associated with a switch to the
defined contribution plan: higher earnings, younger, female, lower tenured, white and nonunion.
Papke (2004) focuses on the impact of vesting and tenure in her study, noting that age and salary
are perhaps of lesser importance in the decision to switch from the defined benefit plan to the
defined contribution plan. She finds that the biggest demand for the defined contribution plan
comes from workers eligible to retire in their 50s and suggests that having a balance to transfer
to the defined contribution plan is positively correlated with a switch.
The remaining research presented in Table 3 above is based on survey work. When asked about
their preferences, DeArmond and Goldhaber (2010) find that more teachers preferred an
additional 10 percent of salary contributed to a defined contribution plan instead of a defined
© 2014 Society of Actuaries, All Rights Reserved Behavioral Research Associates 22 | P a g e
benefit plan. Lesser-experienced employees as well as those who reported participating in a
hybrid plan or not knowing their plan type were more likely to report this preference. The
authors note that the market’s favorable performance in the spring of 2006 may have been a
factor in respondents’ preferences. However, as would be expected, low discounters preferred
the additional contribution to a defined benefit plan. With respect to the respondent group’s plan
knowledge, a high percentage of them in the traditional defined benefit plans could correctly
describe their plan, but less than half of those in the hybrid plan could.
Similar to Benartzi and Thaler (1999), who found that 58 percent of faculty spent less than an
hour deciding how much to contribute to their retirement plan and how to invest their
contributions, Clark, Harper and Pitts (1997) found that 29 percent of respondents “did not give
much consideration to the choice of my primary retirement plan” and only 22 percent said that
they put a great deal of thought into their decision. In addition, nearly 30 percent of respondents
didn’t consult with anyone when selecting a plan. They also found that more than 40 percent
were unaware their decision was irrevocable.
In summary, we see that a significant portion of individuals don’t put much thought into their
retirement plan choice and simply accept the default, not realizing the choice is irrevocable.
Passive as well as active choices frequently appear to be suboptimal, and behaviorists attribute
these suboptimal choices to anomalies that reappear throughout employees’ retirement planning
cycle.
Retirement Decisions During the Working Years
Nearly 70 percent of all employees work for entities that provide retirement benefits, a statistic
that masks wide variation between the percentage of private-industry employees who have
access (65 percent) and the percentage of state and local government workers who do (89
percent) (U.S. Department of Labor, 2012).27 Over the last 30 years, and in particular since
subsection IRS 401(k) of the Internal Revenue Code became effective, defined contribution
retirement plan benefits have become much more prevalent, and are available to 55 percent of
civilian workers (59 percent in private industry and 31 percent in state and local governments) as
of March 2012. Over the same period, defined benefit plans have become less common,
particularly in private industry, and are now only accessible to 29 percent of workers (19 percent
in private industry and 83 percent in state and local governments) (U.S. Department of Labor
2012). By comparison, in 1975, 70.8 percent of private-industry active workers were covered by
defined benefit plans (U.S. Department of Labor, Employee Benefit Security Administration,
2013).
The retirement planning decisions employees face within these two types of retirement benefit
programs are vastly different, particularly during the working years, or what is often referred to
as the accumulation phase. Employees covered by defined benefit plans make relatively few
decisions and shoulder little responsibility for funding their retirements. The employer is
(generally) fully responsible and bears all risk of funding and investing assets of the plan from
27 A number of other disparities are noted. For example, access varies by firm size, as well as the position, union membership,
employment status (full or part time) and income level of the employee (U.S. Department of Labor, 2012).
© 2014 Society of Actuaries, All Rights Reserved Behavioral Research Associates 23 | P a g e
which retirement benefits will be paid. The primary risks employees face in this environment
include the risk of future benefit reductions (either through plan design changes or employer
insolvency) and involuntary separation prior to normal retirement age.
Within the majority of defined contribution plans, employees and employers have markedly
different roles. Employees are fully responsible for funding their retirement years and bear all
related risks. Employers, often with assistance from advisers and consultants, are “choice
architects”28 responsible for developing plan decision-making contexts that so significantly
impact employees’ retirement outcomes, as more fully discussed below. Employers decide the
action required for employees to participate in the plan. They select the investment choice set
from which employees may pick their retirement funds. They determine what steps are required
for employees to develop and manage well-diversified portfolios. And, they decide whether
employees will have preretirement access to their retirement assets via loans and/or hardship
withdrawals. Finally, employers determine what payout options are available. Employees’
retirement choices are executed within these employer-defined constraints, and the increased
prevalence of defined contribution plans as the sole workplace retirement benefit increases the
need for optimal decision-making.
In this section, the potential decisions employees face within employer-sponsored retirement
plans are discussed. These include participation, contribution, investment and withdrawal
choices, virtually all of which are relevant for many types of retirement plans—most notably
salary-deferral-type plans such as 401(k) and 403(b) plans. Differences between plan-type
contexts are noted. Observed behaviors and, where available, individual characteristics of
decision-makers are covered. Behavioral anomalies are highlighted, as are explanations posited
by researchers.
The Participation Decision
Across the board, the U.S. Department of Labor reports that approximately 54 percent of civilian
workers participate in workplace retirement benefit programs, for an overall blended “take-up
rate” of 79 percent. Take-up rates in private industry and state and local government are 75 and
95 percent, respectively (U.S. Department of Labor, 2012).
Other data show lower rates of access and participation (Copeland, 2012; Purcell, 2009a).29 For
example, Copeland (2012) analyzes March 2011 Current Population Survey (CPS) data and
reports statistics for three main groups of employees: all workers, which includes
unincorporated, self-employed individuals; “wage and salary” workers between the ages of 21
and 64; and “full-time, full-year” employees of the same age.30 Table 5 below reports these
findings.
28 A “choice architect,” as defined by Thaler and Sunstein (2008), is one with “the responsibility for organizing the context in
which people make decisions.” They further note the parallels between a traditional architect and a choice architect to make the
point that a neutral design does not exist. 29 For a discussion of pension coverage using different data sets, see “Estimating Pension Coverage Using Different Data Sets”
by G. Sanzenbacher (2006). 30 CPS data comprise survey results from a representative sample of 97,000 households, collected annually by the Census
Bureau. Each March, two questions related to workplace retirement benefits are included in survey. Both Copeland (2012, pp. 7
© 2014 Society of Actuaries, All Rights Reserved Behavioral Research Associates 24 | P a g e
Table 5. Various work forces working for an employer that sponsored a retirement plan
and the percentage of workers participating in a plan, 2011
All
Workers
Wage and Salary
Workers
Ages 21–64
Private Sector
Wage and Salary
Workers
Ages 21–64
Public Sector
Wage and
Salary Workers
Ages 21–64
Full-Time,
Full-Year
Wage and
Salary Workers
Ages 21–64
(millions)
Worker
Category
Total
153.7 128.6 108.3 20.4 91.0
Works for
an employer
sponsoring a
plan
75.2 69.3 52.8 16.5 55.4
Participating
in a plan
61.0
57.4
42.5
14.9
48.8
(percentage)
Worker
Category
Total
100.0 100.0 100.0 100.0 100.0
Works for
an employer
sponsoring a
plan
48.9 53.9 48.8 81.1 60.8
Participating
in a plan
39.7
44.6
39.2
73.2
53.7
Source: Employee Benefit Research Institute estimates from the 2012 March Current Population Survey
Note. From, “Employment-Based Retirement Plan Participation: Geographic Differences and Trends, 2011,” by C.
Copeland, 2012, EBRI Issue Brief 378. Employee Benefit Research Institute.
In contrast to U.S. Department of Labor data that include information about plan type as well as
other employer-level information, the CPS data do not include this information but instead
provide more information about the individuals surveyed, enabling researchers to identify
differences in coverage among various demographic segments (Copeland, 2012). In his analysis
of the March 2012 data, Copeland finds that older, married or divorced, white, more educated
(schooled), male, full-time workers in better health with higher incomes and employer-sponsored
health insurance are more likely to participate in an employer-sponsored retirement plan. In the
population of all workers, men are more likely to participate, but full-time women who worked a
full year were more likely to participate than their male counterparts.
and 8) and Purcell (2009a, pp. 13-15) address the difference between U.S. Department of Labor data (which are from employer
survey responses) and CPS data. For trend information, see pp. 27-29 of Copeland (2012).
© 2014 Society of Actuaries, All Rights Reserved Behavioral Research Associates 25 | P a g e
Participation rates vary by plan type as indicated in Table 6 below, potentially attributable to the
opposing manner in which employees come to be participants in plans. Historically, a significant
difference between the decision-making contexts of defined benefit and defined contribution
plans related to the action required to participate. Generally, this difference persists, but to a
lesser extent than it did in the past when eligible workers automatically became participants in an
entity-sponsored defined benefit plan and only became participants in a defined contribution plan
if they had actively enrolled in the plan. Now, over 40 percent of companies automatically enroll
participants in defined contribution plans (Plan Sponsor Council of America [PSCA], 2011), a
context change discussed below that has important implications for defined contribution plan
take-up rates.
Table 6. Plan participation (take-up) rates by employment sector and type of plan
Defined Benefit Defined Contribution Total
All Civilians 26% (91%) 37% (68%) 54% (79%)
Private Industry 17% (89%) 41% (70%) 48% (75%)
State and Local Government 78 % (94%) 15% (48%) 84%(95%)
U.S. Department of Labor (2012)
Note: Participation percentages represent percentage of workforce indicated participating in plan type and not
percentage of eligible employees.
Note: From “National Compensation Survey,” 2012. United States Department of Labor.
In the private sector, eligible workers generally still become automatic participants in company-
sponsored defined benefit plans, but in the public sector, some states offer alternatives to a
primary defined benefit plan, and employees may choose their preferred plan.31 See “Plan Type
Preferences” for a discussion of selected research on employee choices in this context.
Within private-industry defined contribution plans, a significant, but declining, portion of
participants must take affirmative action to participate as noted above.32 Participation must often
be effected via the web or an automated phone line, but some firms permit the use of
representative-assisted enrollment or paper forms. Within the public sector, defined contribution
plans tend to be supplemental plans due to the high prevalence of defined benefit plans
(Wiatrowski, 2009).33 Participation is typically voluntary and similar to their private-industry
counterparts, workers must sign up to participate. However, in plans with employer
contributions, participation, including employee contribution, may be required as a condition of
employment.
31 These states include Colorado, Florida, Indiana, Montana, North Dakota, Ohio, South Carolina, Utah and Washington (Snell,
2012). 32 Participation is automatic (i.e., no employee action is required) in some types of defined contribution plans such as money-
purchase plans, profit-sharing plans and certain stock plans. In addition, no affirmative action is required to participate in some
savings and thrift plans such as automatic enrollment 401(k) plans. Approximately 19 percent of private-industry workers
participating in defined contribution plans (which in this case do not include 401(k) and other salary-deferral plans, which are
classified as savings and thrift plans in U.S. Department of Labor data) are participants in money-purchase plans, which are fully
funded by employer contributions (U.S. Department of Labor, 2010). Twenty-one percent of private-industry workers in savings
and thrift plans are in plans with an automatic enrollment feature (U.S. Department of Labor, 2010). 33 Wiatrowski (2009) notes that only 5 percent of public-sector employees participate exclusively in a defined contribution plan.
© 2014 Society of Actuaries, All Rights Reserved Behavioral Research Associates 26 | P a g e
The decline in the provision of retirement benefits that provide lifelong monthly income to
private-industry employees intensifies the importance of planning for retirement-income needs.
For many, participating in an employer’s defined contribution retirement plan is the most
convenient way to begin making those preparations, and researchers are keen to assess who is
doing so, and what influences their decision-making. In the remainder of this section, the
individual characteristics of who participates in workplace retirement plans are reported. After
discussion of the extent to which employees are contributing to workplace retirement programs,
plan features, social norms and other behavioral factors that impact the retirement plan
participation and contribution decisions are covered.
In their efforts to identify the individual characteristics of who is choosing to participate in
private-industry defined contribution plans, researchers have analyzed a variety of data sources
including CPS data (Andrews, 1992; Bassett, Fleming, & Rodrigues, 1998; Even & Macpherson,
2005), Health and Retirement Study (HRS) data (Papke, 2003a and 2003b), SCF data (Munnell,
Sunden, & Taylor, 2001/2002; Munnell et al., 2009), Survey of Income Program Participation
(SIPP) data (Smith, Johnson, & Muller, 2004), plan-specific records (Clark & Schieber, 1998;
Kusko, Poterba, & Wilcox, 1998; Agnew, 2006a; Huberman, Iyengar & Jiang, 2007) and other,
smaller-scale survey results (Bernheim & Garrett, 2003).
Where these demographic variables have been observed and measured, the researchers have
found positive associations between age, income and tenure, and the probability of participating.
Kusko, Poterba and Wilcox (1998) find age is more important at lower-income levels, and Smith,
Johnson and Muller (2004) find that increases with earnings disappear at higher-income levels
and that increases with age occur up to about age 55.
Other positively related factors include education and home ownership (Andrews, 1992; Bassett,
Fleming & Rodriques, 1998; Papke, 2003a; Smith, Johnson & Muller, 2004), although Munnell,
Sunden & Taylor (2001/2002) and Bernheim & Garrett (2003) find no significant effect from
education. Gender effects are not clear. When gender effects are found, women are 4 (Agnew
2006a) to 6.5 (Huberman, Iyengar & Jiang, 2007) percent more likely than men to participate.34
However, some researchers find no significant relationship between gender and participation
(Bernheim & Garrett, 2003; Smith, Johnson & Muller, 2004).
While Munnell, Sunden and Taylor (2001/2002) find a positive relationship between net worth
and participation, Papke (2003a) finds no significant relationship between the two. Having a
short time horizon (Munnell, Sunden & Taylor, 2001/2002) and being married (Bassett, Fleming
& Rodriques, 1998; Smith, Johnson & Muller, 2004) are each negatively related to participation
likelihood. However, in Munnell et al.’s analysis of 2007 SCF data, they find no significant
relationship between having a short time horizon (defined as a planning horizon of four years or
less) and participation (2009). The authors attribute this change to the increase in automatic
enrollment, where more people without a preference for saving become plan participants.
Using longitudinal data, Smith, Johnson and Muller (2004) analyzed the effects of various life
changes on plan participation. They find no impact of a change in marital status. Participation
rates declined with the number of children under the age of 18, but the probability of
34 Papke (2003a) finds that single females are 5 percent more likely to participate than married males.
© 2014 Society of Actuaries, All Rights Reserved Behavioral Research Associates 27 | P a g e
participating increased in the year of or the year following the birth of a child. Work-limiting
health problems had no impact on one’s participation, but spousal health problems corresponded
with an increase in the probability of participation. No earnings for a year reduced the probability
of participating (potentially due to eligibility requirements). Participation likelihood increased
when a spouse changed or returned to work.
Two of the aforementioned studies show that the decision to participate in a 401(k) plan may be
a signal of whether the worker considers the employment relationship to be short term. Even and
MacPherson (2005) find that for workers with less than three years of tenure, the probability of a
job change is 14 percentage points higher for workers who choose not to participate in the 401(k)
plan.35 Kusko, Poterba and Wilcox (1998) report similar results. In their research, they find an
average first-year participation rate of 50 percent among new hires, but among those who left,
the participation rate was a mere 6.5 percent.
The Contribution Decision
The retirement plan contribution decision is most relevant to defined contribution plans, and
more specifically salary-deferral plans. Although some defined benefit plans require employee
contributions (typically public-sector plans), the percentage of one’s salary that must be
contributed is typically specified.36 This is not the case with most defined contribution plans,
particularly in private industry. To participate in salary-deferral plans, the most popular type of
defined contribution plan that includes both 401(k) and 403(b) plans, employees must decide
how much of their paycheck to divert to the plan. It is not an easy decision, especially when one
considers all of the inputs that a fully rational decision would require.
Based on SIPP data collected in early 2012, conditional on participation, the average
contribution to salary-deferral plans was 6.7 percent, which represents a decline from the 7.4
percent reported in 2009 (Copeland, 2013).37 Approximately 53 percent of respondents
contributed 5 percent or less, and nearly 25 percent of respondents contributed between 5 and 10
percent. The remaining 22 percent contributed 10 percent or more (Copeland, 2013).
Similar to the individual characteristics positively associated with plan participation, researchers
have generally found that higher contribution rates are associated with increases in age, income
and tenure.38 For example, Holden and VanDerhei (2001), who analyze contribution behavior of
1.7 million 401(k) plan participants drawn from the EBRI/Investment Company Institute (ICI)
Participant-Directed Retirement Plan Data Collection Project, estimate that contribution rates
increase by .06 percentage point for each additional year of age for participants in their mid-40s
or younger. For older participants, the increase is estimated at .07 percentage point per additional
35 This compares to an overall probability of job change for workers with less than three years of tenure of 27 percent (Even &
Macpherson, 2005). 36 Based on U.S. Department of Labor data, 83 percent of public-sector workers participating in defined benefit plans are required
to contribute, on average, 6.6 percent (2012). 37 This finding, based on self-reported data, compares to 7.0 percent and 7.3 percent in 2007, which is based on 2012 Vanguard
recordkeeping data for over 1,600 plans and 3 million participants. 38 For example, see Andrews (1992); Xiao (1997); Clark and Schieber (1998); Kusko, Poterba and Wilcox (1998); Holden and
VanDerhei (2001); Papke (2003a); K. Smith, Johnson and Muller (2004); and Huberman, Iyengar and Jiang (2007).
© 2014 Society of Actuaries, All Rights Reserved Behavioral Research Associates 28 | P a g e
year (Holden & VanDerhei, 2001). Smith, Johnson and Muller (2004) estimate that the increases
begin after about age 33, and, using 1995 SCF data, Xiao (1997) estimates that the increases
occur until about age 45 or 46 before falling. However, Munnell, Sunden and Taylor
(2001/2002) estimate there is no significant influence from age.
Holden and VanDerhei (2001) and others also report that higher salaries and increased tenure are
associated with higher levels of contributions up to a point. With respect to income levels, it is
likely that regulatory and plan limits curtail additional tax-deferred contributions.39 The positive
effects of tenure appear to fall after about 18 years on the job (Holden & VanDerhei, 2001).
Certain research has also suggested some gender effects on contribution levels. Huberman,
Iyengar and Jiang (2007), Papke (2003a) and VanDerhei and Copeland (2001) find that women
contribute more than men, but others find no difference between the two. Munnell, Sunden and
Taylor (2001/2002) also find a significant effect associated with having a short planning horizon.
They suggest that a planning horizon of less than five years predicts a contribution rate that is 1.2
percentage points lower. The effect of planning horizon in Munnell et al.’s (2009) analysis of
SCF 2007 data is minimal and insignificant.
Factors Affecting Participation and Contribution Levels
A review of relevant literature reveals several factors that affect employees’ retirement plan
decision-making, often in surprising ways, evidencing employees’ irrational decision-making
tendencies. The steps an employee must take to start contributing to a retirement matter, as do
other plan features such as the existence of saving-rate-increase programs, employer matching
contributions and loan provisions. Research results also indicate that the nature of the investment
menu offered to employees has an effect on participation likelihood. In this section, the role of
these plan features as well as social norms and other decision-making heuristics and biases are
discussed.
Enrollment-Related Features
Automatic Enrollment
Over the last 30 years, sponsors of defined contribution retirement plans have increasingly
“reframed” the participation decision from requiring action to join the plan to requiring action to
avoid it.40 Plans that require action to avoid joining are automatic enrollment plans. Since no
action is required for participant enrollment into the plan, employers must select an initial
contribution rate and make an investment choice that will be used for all automatically enrolled
participants. These are referred to as default choices. Employees are free to make different
choices, but if they do not, they will become plan participants, contributing at the default rate and
investing in the default investment selected by the plan sponsor.
39 Further, Holden and VanDerhei (2001) note a nonlinear relationship between income and contribution rates, reporting greater
influence from salary increases at higher levels of earnings. 40 That automatic enrollment is a reframing of the enrollment decision is set forth by Madrian and Shea (2001a) and Mitchell and
Utkus (2004).
© 2014 Society of Actuaries, All Rights Reserved Behavioral Research Associates 29 | P a g e
The percentage of defined contribution plans with an automatic enrollment feature has
dramatically increased over last decade, and as of 2010, about 42 percent of 401(k) plans offer an
automatic enrollment feature (PSCA, 2011). (See Figure 2 below.) U.S. Department of Labor
National Compensation Survey (2010) data show that approximately 20 percent of private-
industry participants in savings and thrift plans have access to an automatic enrollment feature.
Figure 2. Percentage of 401(k) plans with automatic enrollment by plan size (number of
participants), 1999–2010
Source: Profit Sharing/401(k) Council of America, Annual Survey of Profit Sharing and 401(k) Plans, Chicago:
Profit Sharing/401(k) Council of America, 2000-2011.
The reframing of the participation decision has been unarguably successful in increasing plan
participation rates, even at higher default contribution rates. In the figures below, results from
automatic enrollment implementations at four companies studied by Choi, Laibson, Madrian and
Metrick (2006) are presented. The details of each company’s enrollment process changes are
provided below each figure showing average participation rates by tenure for relevant cohorts.
The positive effect of automatic enrollment on participation rates is obvious from these figures.
After six months of tenure, participation rates are 50 to 67 percentage points higher under
automatic enrollment (Choi et al., 2006). Since the probability of participation increases with
tenure when automatic enrollment is not used, the participation rate differential declines over
time and at 36 months is 20 to 34 percentage points (Choi et al., 2006). In an environment of
increased mobility, these differences in participation rates could easily result in sizable gaps
between retirement assets generated under automatic enrollment and the level of assets
accumulated under a traditional enrollment model.
0%
10%
20%
30%
40%
50%
60%
70%
1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010
All 1-49 50-199 200-999 1,000-4,999 5,000+
© 2014 Society of Actuaries, All Rights Reserved Behavioral Research Associates 30 | P a g e
Figure 3. Company B: 401(k) participation by tenure
Automatic enrollment (2 percent contribution rate and a stable value fund as the default investment) was
implemented at the beginning of 1997 for new employees only and subsequently dropped four years later.
Figure 4. Company C: 401(k) participation by tenure
Automatic enrollment (3 percent contribution rate and a money market fund as the default investment) was
implemented in April 1998 for new employees only. Three years later, the default investment was prospectively
changed to a lifestyle fund. At the same time, contribution rates of participants with one year of tenure who
remained at the initial default rate of 3 percent were prospectively increased to 6 percent. Therefore, beginning in
May 2001, employees still contributing 3 percent after a year of tenure automatically began contributing 6 percent.
Figure 5. Company D: 401(k) participation by tenure for employees age 40+ at hire
© 2014 Society of Actuaries, All Rights Reserved Behavioral Research Associates 31 | P a g e
Automatic enrollment (3 percent contribution rate and a stable value fund as the default investment) was
implemented at the beginning of 1998 for new employees and, subsequently, automatically enrolled eligible
nonparticipants. Three years later, the default contribution rate was increased to 4 percent.
Figure 6. Company H: 401(k) participation by tenure
Automatic enrollment (6 percent contribution rate and a balanced fund as the default investment) was implemented
at the beginning of 2001 for new employees only.
Note. Figures 3 through 6 are from "Saving for Retirement on the Path of Least Resistance,” by J.J. Choi, D.
Laibson, B.C. Madrian, & A. Metrick. In Ed McCaffrey and Joel Slemrod, eds., Behavioral Public Finance: Toward
a New Agenda. New York, NY: Russell Sage Foundation, 2006: 304-351. © 2006 Russell Sage Foundation.
Reprinted with permission.
Automatic enrollment’s positive impact on plan participation comes at the expense of lower
contribution rates partially as a result of the very behavioral tendency that enables its success—
inertia.41 The researchers find that while automatically enrolled participants have been nudged in
a presumably better direction (toward plan participation), many of them are proceeding with
lower contribution rates than if they had actively chosen to participate on their own (Madrian &
Shea, 2001a). In the 3 percent automatic enrollment they studied at one large company, Madrian
and Shea (2001a) note raw and regression-adjusted reductions in average contribution rates of
2.9 and 2.2 percentage points, respectively, that appear across virtually all demographic
41 Note that the selection of higher default rates could possibly help overcome this situation.
© 2014 Society of Actuaries, All Rights Reserved Behavioral Research Associates 32 | P a g e
segments. A Vanguard analysis of employees hired between January 1, 2004, and September 30,
2006, in 527 voluntary and 48 automatic enrollment plans showed similar results. Conditional on
plan participation, the average contribution rate of employees hired under voluntary enrollment
was 6.1 percent, compared to an average of 4.2 percent for automatically enrolled participants
(Nessmith, Utkus & Young, 2007). See Table 7 below.
Table 7. Voluntary versus automatic enrollment for new hires Hired under voluntary enrollment Hired under automatic enrollment
Eligible employees 319,002 18,544
Participation rate 45% 86%
Contribution Rates
Employee average 2.8% 3.6%
Employee median 0.0% 2.6%
Participant average 6.1% 4.2%
Participant median 5.0% 2.9%
Note: For employees hired between Jan. 1, 2004, and Sep. 30, 2006, as of Dec. 31, 2006
Note. From “Measuring the Effectiveness of Automatic Enrollment,” by W.E. Nessmith, S.P. Utkus, and J.A.
Young, 2007, Vangard Center for Retirement Research.
In addition, the tendency to passively accept the default contribution rates persists over long
periods of time for a significant percentage of the automatically enrolled population. Even after
three years, between 29 and 48 percent of automatically enrolled participants remain at the
default rate and wholly invested in the default fund (Choi et al., 2006).42 Income appears as the
strongest predictor of moving away from the defaults (Choi, Laibson, Madrian, & Metrick,
2004a). These persistency levels are higher than those found by Nessmith, Utkus and Young
(2007), however, as shown in Figure 7 below.
42 In a discussion with one of the study’s authors, he relayed that the persistence of the default rate and default investment were
about the same.
© 2014 Society of Actuaries, All Rights Reserved Behavioral Research Associates 33 | P a g e
Figure 7. Automatic enrollment over time
Note. From “Measuring the Effectiveness of Automatic Enrollment,” by W.E. Nessmith, S.P. Utkus, and J.A. Young,
2007, Vangard Center for Retirement Research.
Nevertheless, the projected benefits of automatic enrollment are significant. Using data from 225
large plan sponsors, VanDerhei (2010) models projected accumulations for various income
segments and estimates that automatic enrollment may increase age-65 accumulations from .08
times final earnings to 4.96 or 5.33 times final earnings (depending on contribution rate change
assumptions at job change) for the lowest-income quartile worker age 25 to 29, assuming
automatic enrollment provisions similar to those used by large plan sponsors in their data set.
VanDerhei (2010) projects benefits for those in the highest income quartile as well.
Simplified Enrollment
In a traditional enrollment environment without nudges, employees select from what is often a
very wide range of allowable contribution rates and investment options. Although regulatory
requirements (and participant liquidity needs) may constrain the upper limit of the permissible
range of contribution rates, at first blush the range may appear as wide as zero to 100 percent of
compensation. And the menu of investment options in plans has grown over the years; the
average number available now stands at 18, a 50 percent increase since 1999 (PSCA, 2011).
Under “simplified enrollment,” employees are offered a preselected contribution rate and
investment choice. Researchers have found that this type of enrollment process can result in a 10
to 20 percentage-point increase over participation rates in traditional enrollment model plans
(Beshears, Choi, Laibson, & Madrian, 2013). Similar to automatic enrollment, the researchers
see participation and default rate persistency. After four years, 90 percent of simplified enrollees
are still in the plan (even seasoned employee enrollees), and between 37 and 63 percent remain
at the initial default rate (Beshears et al., 2013).
© 2014 Society of Actuaries, All Rights Reserved Behavioral Research Associates 34 | P a g e
Required Active Decision-Making
Under what researchers call the “required active decision-making” enrollment process,
employees must make (and communicate) a decision about plan participation (even if the
decision is to forgo enrollment) within a limited, specified, but loosely enforced, time frame. In
other words, inertia is thwarted. Researchers implementing this enrollment method found that
enrollment rates were 28 percent higher than they were under a traditional model. Further, they
note that requiring active participation decisions may be optimal for procrastination-prone
workforces with a wide range of optimal savings rates (Carroll, Choi, Laibson, Madrian, &
Metrick, 2005).
It is illogical that so many more would decide to save (and exactly the same amount as the
default rate) for retirement simply because of these differences in the enrollment process.
Researchers posit that automatic enrollment, quick enrollment and required active decisions are
effective because they address (if not capitalize on) decision-making tendencies that can, and
often do, hinder economically ideal retirement preparation. While two of these tendencies (loss
aversion and self-control problems) are discussed in “Automatic Deferral Increase Programs,”
here we describe others that are at play: status quo bias, procrastination and framing effects.
In a series of experiments, Samuelson and Zeckhauser (1988) discover an attachment to the
status quo (i.e., “status quo bias”) that is particularly strong in cases where there are no strong
preferences and when the choice set is large. Certainly, the choice set of most salary-deferral-
type plans is huge, and it is easy to imagine that many would have no particular preferences. A
status quo bias may stem from procrastination caused by the costs (search or transactional)
associated with moving away from the status quo and it may also be caused by the self-control
problem discussed below. Researchers (Madrian & Shea, 2001a) suggest procrastination indeed
may help explain some of the difference between participation rates in standard, automatic and
quick enrollment plans. In the automatic and quick enrollment environments, the participation
decision is reduced to a binary one. There is no longer the requirement to think about how much
to save or which funds to select.
Automatic enrollment is often referred to as a “reframing” of the participation decision from one
that requires action to join to one that requires action to not join (Madrian & Shea, 2001a;
Mitchell & Utkus, 2004). The way options are presented shouldn’t affect choices, but there are
numerous examples across many domains illustrating that it does. The classic example in
behavioral economics demonstrates the effect of framing options in terms of gains or losses.
Tversky and Kahneman (1981) show that when a riskless and a risky option are described in loss
terms, preferences are reversed from when the options are described in terms of their gains.
While the status quo bias and procrastination may explain the persistency of automatic
enrollment default choices, researchers suggest other forces may also be affecting the continued
acceptance of plan defaults. McKenzie, Liersh and Finkelstein (2006) suggest employees
perceive the default choices (including the default choice of enrollment) as implicit advice. In
their lab experiment, they find that when subjects were told they would be automatically enrolled
in the company’s retirement plan, they were more likely (than those who weren’t automatically
© 2014 Society of Actuaries, All Rights Reserved Behavioral Research Associates 35 | P a g e
enrolled) to believe the human resources staff think employees should enroll (80 versus 11
percent).
Madrian and Shea (2001a) further offer that once they have been “endowed” with plan
participation, automatically enrolled participants may value their participation much more than
they would value discontinuing it. Thaler coined this anomaly stemming from the concept of loss
aversion the “endowment effect” (Thaler, 1980).
Automatic Deferral Increase Programs
Automatic deferral increase programs, conceived by Benartzi and Thaler (2004) and coined
“SMarT” for “Save More Tomorrow” are another plan design feature that beneficially exploits
human (as opposed to fully rational) decision-making tendencies to improve retirement savings
rates over time. As it was conceived, employees precommit to increasing savings rates in the
future when they receive pay raises.43
In first implementation of the Save More Tomorrow program, 78 percent signed up for the
service, and 80 percent continued with it through the fourth pay raise. The average savings rates
for SMarT participants increased from 3.5 to 13.6 percent over the 40-month period covered
(Benartzi & Thaler, 2004).44
Benartzi and Thaler (2004) designed the program with behavioral anomalies in mind. In addition
to simplifying the savings decision and taking advantage of procrastination and inertia, both of
which have been discussed above, SMarT addresses problems of self-control and loss aversion.
It also addresses what is known as “money illusion” by offering employees a way to commit to
having better self-control when a pay raise is given, thereby relieving a perceived loss even if the
raise is an illusion because it is below what would be necessary to keep pace with inflation
(Benartzi & Thaler, 2004).
Referring to McIntosh (1969), Thaler and Shefrin (1981) rationalize their use of a two-self model
in their economic theory of self-control, noting that the idea of self-control is paradoxical
without it. They suggest a farsighted planning-self and a shortsighted doer-self. The farsighted
planning-self would like to save more for a comfortable retirement but the shortsighted doer-self
would much rather spend more today. Put differently, individuals are said to have time-
inconsistent preferences related to higher levels of impatience in the short term than in the longer
term.45 SMarT provides a way for the shortsighted doer-self and the farsighted planning-self to
exist in harmony. One can spend today and yet at the same time commit to save more in the
future.
Because saving generally requires a reduction in current consumption, a sense of loss may be
experienced, and behaviorists have discovered that the pain of a loss is about two to two and a
43 Due to administrative limitations, the savings-rate increases are often not synchronized with pay raises. 44 An adviser who met individually with most eligible employees conducted the first implementation and this personalized
attention may have had an effect. The results of two other implementations, one of which was conducted entirely via mail, had
lower participation rates. 45 The term “hyperbolic discounting” is also used to describe time-inconsistent preferences, or present-based biases. For
additional information, see Frederick, Loewenstein and O’Donoghue (2002).
© 2014 Society of Actuaries, All Rights Reserved Behavioral Research Associates 36 | P a g e
half times the pleasure of a gain (Tversky & Kahneman, 1992; Kahneman, Knetsch, & Thaler,
1990). Because savings-rate increases occur when a pay raise is received, SMarT participants
avoid a sense of loss. And finally, SMarT takes advantage of the “money illusion” in the event
that a raise is nominal rather than real. (“Money illusion” occurs because people tend to think in
nominal rather than real terms [Fisher, 1928.])
Employer Matching Contributions
Approximately 89 percent of private-sector 401(k) plans with more than 100 participants offer
employer matching contributions (Soto & Butrica, 2009). Since employer matching contributions
provide additional economic incentive for employees to contribute to the plan, the widely
accepted rationale for the provision of matching contributions is their ability to encourage
employee participation and motivate higher contribution levels. But do they? Although most
researchers have found that the existence of an employer matching contribution increases plan
participation, the effect on employee contribution rates is more ambiguous, as shown in Table 8
below. At least some of the varying results are believed to stem from incomplete data sets.46
Table 8. Research on employer matching contributions
Study and Data Description Effect on Participation Effect on Employee Contributions
Studies Using Actual Behavioral Data
Kusko, Poterba and Wilcox
(1998)
Plan records from 1988 through
1991 for 12,000 employees
from one manufacturing firm
Minimal, except that when match
rate increased from 63 percent to
150 percent, 63 percent of prior
nonparticipants joined
Minimally positive
Clark and Schieber (1998)
1994 participant and plan data
from 19 firms (all plans offered
employer matching
contributions)
Positive effect from higher match
rates
Positive effect from higher match rates
(model estimated 50 to 75 percent match
raises contribution level by .8 percentage
point, compared to 25 percent match, and
100 percent match raises contribution levels
by 2 percentage points higher than 25
percent match)
VanDerhei and Copeland
(2001)
Demographic and contribution
behavior for over 160,000
participants between the ages of
21 and 64, with $10,000 in
earnings and a contribution in
1998
Positive
Negative for match rate and total potential
match
Match threshold a significant factor
46 For a brief overview, see Engelhardt and Kumar (2003).
© 2014 Society of Actuaries, All Rights Reserved Behavioral Research Associates 37 | P a g e
Study and Data Description Effect on Participation Effect on Employee Contributions
Holden and VanDerhei (2001)
Demographic and contribution
behavior from 1999 for
approximately 1 million
participants
Not evaluated
Minimally negative effect from increased
match rate
Positive effect from increased match
threshold
Choi, Laibson, Madrian and
Metrick (2002)
Participant and plan data from
1998 to 2000 for two plans (one
with approximately 10,000
employees and the other with
40,000)
Positive (versus no match)
No effect from increase in match
threshold
Positive related to threshold
Positive related to increase from zero to 25
percent match
Mitchell, Utkus and Yang
(2007)
Participant and plan data from
2001 for approximately 500
plans covering nearly 750,000
participants
Minimally positive (estimate that
an additional 10 percent would
participate in presence of match as
compared to no match),
particularly for nonhighly
compensated employees
See Table 9 below.
Positive
As match rates increase (particularly on the
first 3 percent of pay), effect becomes
negative
See Table 9 below.
Huberman, Iyengar and Jiang
(2007)
Participant and plan data from
650 plans covering nearly
800,000 eligible employees
Positive, strongest effect on lower-
income participants
Positive
Middle-income participants contribute less
when the match is generous
Studies Using Survey and Other Data
Andrews (1992)
May 1988 CPS data
Positive
Higher match rates are negatively related to
contribution levels
Bassett, Fleming and
Rodrigues (1998)
1993 CPS data
Estimated participation increase of
10 percentage points in presence
of matches
Higher matches do not relate to
higher participation
No effect
Munnell, Sunden and Taylor
© 2014 Society of Actuaries, All Rights Reserved Behavioral Research Associates 38 | P a g e
Study and Data Description Effect on Participation Effect on Employee Contributions
(2001/2002)
1998 SCF data
Not evaluated
Positive, presence of match increases
predicted contribution rate by .7 percentage
point
Larger matches have a negative impact
Papke (1995)
1986 and 1987 Form 5500 data
Positive (estimated 17.4
percentage point increase in
participation from an increase in
match rate from zero to 100
percent)
Positive up to 80 percent, but beyond that,
negative effect (lower employee
contributions than in plans without matches)
Papke and Poterba (1995)
1986 and 1990 survey data
from 43 firms
Positive
Positive impact from higher
matches (estimated 26 percentage
point increase in participation
from an increase in match rate
from zero to 100 percent)
Weak positive relationship
Even and Macpherson (2005)
1993 pension supplement to
CPS data
Positive under multiple models
Not evaluated
Engelhardt and Kumar (2007)
1992 HRS data linked to Social
Security and Internal Revenue
Service earnings records for
1,042 subjects
Positive
A 25 percent match is estimated to
increase participation by 5
percentage points
Positive, but inelastic
Xiao 1997
1995 Survey of Consumer
Finances
Positive to a point (when match reached
$7,076 in one model and 22 percent in
another model) and then negative
U.S. General Accounting
Office 1997
1992 SCF data and 1992 Form
5500 research data base
Positive, estimate that matching
increases participation by 20
percentage points
Estimate that matches increase contribution
levels 10 to 24 percent, depending on match
rate
© 2014 Society of Actuaries, All Rights Reserved Behavioral Research Associates 39 | P a g e
Study and Data Description Effect on Participation Effect on Employee Contributions
Based on regressions, no positive
relationship between match rates and
predicted contribution rates appears
Threshold levels not evaluated
Munnell et al. 2009
2007 SCF data
Positive
Positive for matches up to 50 percent, then
negative
Threshold levels not evaluated
Dworak-Fisher 2010
2002 and 2003 U.S.
Department of Labor microdata
exclusively including plans
with an employer match (6
percent of plans were automatic
enrollment plans)
Little to no effect on participation
of lower-income segment
Small positive but significant
effect on middle-income segment
Not evaluated
Table 9. Forecasted value of non-highly compensated employee participation and saving
rates based on match design, with all other independent variables estimated at means
Panel A: Forecasted Participation Rates
Base participation rate (no match, no loan): 63%
Base participation rate (no match, with loan): 64%
Match Rates
Match tier $0.25 on the dollar $0.50 on the dollar $1.00 on the dollar
3% 69% 71% 76%
4% 69% 72% 77%
5% 70% 73% 78%
6% 71% 73% 78%
Panel B: Forecasted Contribution Rates
Base savings rate (no match, no loan): 6.1%
Base savings rate (no match, with loan): 6.7%
Match Rates
Match tier $0.25 on the dollar $0.50 on the dollar $1.00 on the dollar
3% 6.6% 6.5% 6.3%
4% 6.6% 6.5% 6.3%
5% 6.6% 6.5% 6.3%
6% 6.6% 6.5% 6.3%
Mitchell, Utkus and Yang (2007)
© 2014 Society of Actuaries, All Rights Reserved Behavioral Research Associates 40 | P a g e
Note. From “Turning workers into savers? Incentives, liquidity, and choice in 401(k) plan design,” by O.S. Mitchell,
S.P. Utkus, & T. Yang, 2007, National Tax Journal, (60), pp. 469-89.
Choi et al. (2002) are able to overcome some of these data limitations in their study of two large
companies with changes in employer matching contributions. One company instituted a match
and the other increased the match threshold (the upper limit on the percentage of compensation
to be matched). In addition to noting an increase in participation when a match was instituted and
an increase in contribution rates when the threshold was increased, they found that the match
threshold has strong influence on contribution rates. Participant contribution rates “cluster”
around the match threshold.47 (See “Anchoring” below for further discussion.)
The growing popularity of automatic enrollment gives rise to a new look at the role of employer
matching contributions within plans with this feature. Beshears, Choi, Laibson and Madrian
(2007) explore this by analyzing participant behavior in one automatic enrollment plan where the
matching contribution was dropped. They find that participation rates after six months of tenure
were 5 to 6 percentage points lower and that average contribution rates declined by .65 percent
of pay. In addition, they aggregate data from a number of automatic enrollment plans with
varying matching contributions to estimate that a 1 percentage point reduction in the maximum
match available predicts a 1.8 to 3.8 percentage-point reduction in participation levels at six
months of eligibility. They conclude that a modestly positive relationship between match
generosity and automatic enrollment plan participation rates exists.
Despite evidence that employer matching contributions positively affect participation and
contribution rates, we also know that a significant percentage of employees fail to take full
advantage of employer matching contributions.48 However, we do not have conclusive evidence
that employees react irrationally to them. While it may seem that employees should take full
advantage of the match, liquidity constraints may simply prohibit this. However, in one study,
Choi, Laibson and Madrian (2011) find contribution choices that would be difficult to
rationalize. In their study, between 20 and 60 percent of match-eligible participants over the age
of 59 1/2 (virtually all of who were fully vested) fail to maximize the benefit available from their
employer matching contributions.49,50 The researchers surveyed a sample of these individuals in
an attempt to identify potential explanations for subthreshold contribution rates and conclude that
procrastination and low levels of financial knowledge appear to at least partly explain participant
contribution decisions.
Investment-Related Features
The ability to select one’s own investments as well as the number and type of investments from
which participants choose can also have an impact on participation and contribution rates. PSCA
(2011) reports that nearly 98 percent of respondent plans offer participants the ability to choose
47 Clustering around match thresholds is observed by others as well. See Benartzi and Thaler (2007), Engelhardt and Kumar
(2007), Madrian and Shea (2001a), and Kusko, Poterba and Wilcox (1998). 48 For example, Mitchell, Utkus and Yang (2007) estimate that the average workforce misses out on about half of the available
company match. Further, Engelhardt and Kumar (2007) estimate that those who contribute leave 1 percent of pay on the table. 49 Participants over the age of 59 1/2 are of particular interest because they can theoretically withdraw the employer-matching
contributions shortly after they are deposited without penalty. The authors note that for the most part, the participants in their
sample were fully vested and conclude vesting is not an issue. 50 Lower-income participants were less likely to take full advantage of the match, as were men and singles.
© 2014 Society of Actuaries, All Rights Reserved Behavioral Research Associates 41 | P a g e
their own investments for their contributions and over 92 percent allow participants to invest
employer contributions made on their behalf. Papke (2003b) suggests that self-direction of plan
investments is one of the most important determinants of (higher) participation and contribution
rates.51 This is confirmed by Z. Li (2012), who finds that individuals with investment choice in
their defined contribution plans contribute over 3 percentages points more than those without
choice.52 This is disputed by Dworak-Fisher (2010) who finds that “Providing workers with a
choice of how to invest their own contributions has a small but significant, negative association
with participation.” Dworak-Fisher finds no effect on participation from the ability to direct the
investment of employer contributions.
Iyengar, Huberman and Jiang (2004) and Mitchell, Utkus and Yang (2007) find that offering too
much choice can have negative consequences. Iyengar, Huberman and Jiang (2004) estimate that
the probability of participation drops by 1.5 to 2 percent for every 10 funds added to the plan’s
investment menu. Mitchell, Utkus and Yang (2007) refine this line of research and offer that
participation of nonhighly compensated employees peaks at 30 investment options and falls
thereafter. Also, they find that the number of funds available is positively related to the
percentage of highly compensated employees saving the maximum allowable amount.
The composition of the investment options offered to employees also matters. An increase in the
proportion of stock funds reduces participation likelihood among nonhighly compensated
employees, but the presence of company stock as an option increases the probability of
participating in the plan, particularly for lower-income employees (Mitchell, Utkus & Yang,
2007; Huberman, Iyengar & Jiang, 2007). More specifically, Mitchell, Utkus and Yang (2007)
estimate that a 10 percent increase in the number of equity options reduces plan participation for
this group by 1.62 percentage points. In Huberman, Iyengar and Jiang’s (2007) participation
estimation model, the presence of company stock increases participation by 2.4 percent, and the
authors suggest a familiarity bias, as discussed below.
“Choice overload” is used to describe Iyengar and Lepper’s (2000) hypothesis that while choice
may at first seem appealing, large choice sets may be demotivating. Having more alternatives
available to suit one’s preferences would, under many circumstances, be welfare enhancing.
However, Iyengar and Lepper (2000) show there are circumstances where choice overload may
cause many people to decide to make no choice at all, similar to the effect of large investment
menus on plan participation. Rather than sort through a daunting list of investment options, some
will decide to defer a decision, and inertia may keep delayers from ever becoming joiners.
Loan Provisions
It is reasonable to expect that the ability to borrow from one’s 401(k) account might encourage
higher plan participation and contribution levels since such access could relieve concerns about
the loss of liquidity associated with contributing to the plan. Most research does in fact estimate
a positive relationship between a loan provision and contribution rates. The estimated impact of a
51 Interestingly, Benartzi and Thaler (2002) find that 80 percent of participants who expressed the desire to construct their own
plan investments actually preferred another investment portfolio constructed by a managed account service. This is even more
interesting because the portfolios were not identified; they were simply specified by a letter. 52 Z. Li’s work is based on the 1992 HRS wave, and men and lower-income participants are more likely to be affected (2012).
© 2014 Society of Actuaries, All Rights Reserved Behavioral Research Associates 42 | P a g e
loan provision on contribution rates ranges from an increase of about 10 percent among
nonhighly compensated employees (Mitchell, Utkus &Yang, 2007), or .6 percentage point
(Holden & VanDerhei, 2001) to 3 percentage points (Munnell, Sunden & Taylor, 2001/2002;
U.S. General Accounting Office [GAO] 1997). Munnell et al. (2009) find a positive effect of
about 1 to 1.5 percentage points. K. Smith, Johnson and Muller (2004) find no effect on observed
contribution rates but do estimate a higher likelihood of participation when workers can borrow
from the plan. GAO (1997) estimates that a loan provision increases plan participation by 6
percentage points. Xiao (1997) and Dworak-Fisher (2010) find no or an insignificant effect from
the ability to borrow, and Mitchell, Utkus and Yang (2007) find no effect on participation
likelihood among nonhighly compensated employees.
Other Retirement Benefits
It is difficult to draw an overall conclusion about the effect of the presence of other retirement
benefits on 401(k) participation and contribution rates. Earlier work found that the presence of
another retirement benefit program lowered participation probability by 11 to 14 percent
(Andrews, 1992; Bassett, Fleming & Rodriques, 1998; Munnell, Sunden & Taylor, 2001/2002).
Papke (1995) and Mitchell, Utkus and Yang (2007) also find that the availability of other plans
reduces the likelihood of participating in the 401(k) plan. Dworak-Fisher (2010) finds a negative
participation effect associated with the employer’s cost of a defined benefit plan but a positive
effect on participation from the presence of another defined contribution plan. Other researchers
who have found positive participation effects from the existence of other plans include: Papke
and Poterba (1995), Even and Macpherson (2005), GAO (1997), Papke (2003a) and Munnell et
al. (2009). Even and Macpherson (2005) suggests that maybe the participation in another plan is
a proxy for a “strong taste for saving.” Munnell et al. (2009) offer that the positive relationship
between 401(k) and defined benefit plan participation may partially be explained by the
existence of frozen defined benefit plans. Finally, research shows that the presence of other
retirement-plan benefits has either a neutral or positive effect on plan contribution rates.53
Social Norms
Social norms can wield significant power over a wide range of behaviors and choices, including
retirement plan participation and contribution. For example, Duflo and Saez (2002) study
individual data from a large university and find evidence participation and vendor choice may be
influenced by the choices of peers. They further conduct a randomized experiment and find that
the attendance rate of individuals in the same department as recipients of an incentive to attend a
benefits fair was three times that of the attendance rate of individuals in other departments who
did not receive an incentive to attend, suggesting some intradepartmental “spillover effects.”
They also find greater variation in interdepartmental enrollment rates than they find between
treatment and control groups within departments (Duflo & Saez, 2003).
Bailey, Nofsinger and O’Neill (2004) conduct a lab experiment with college students and find a
strong main effect for what the researchers call descriptive norms (telling people what similar
53 Munnell, Sunden and Taylor (2001/2002); Munnell et al. (2009); and Xiao (1997) find no effect. Papke and Poterba (1995);
GAO (1997); Papke (2003a); Mitchell, Utkus and Yang (2007); and Huberman, Iyengar and Jiang (2007) find that participants
with access to another plan contribute more than participants for whom the 401(k) plan is the sole retirement plan.
© 2014 Society of Actuaries, All Rights Reserved Behavioral Research Associates 43 | P a g e
others typically do) and injunctive norms (telling people what experts say they ought to do). On
the other hand, other researchers have found no evidence of peer effects from providing
employees who had opted out of an automatic enrollment plan information about the savings
decisions of their peers (Beshears, Choi, Laibson, Madrian, & Milkman, 2011). In fact, the effect
of providing this information to a group of unionized employees was a larger gap between their
average contribution rate and the published rate of their peers.
Other Observed Heuristics and Biases Affecting Saving Decisions
Thus far, the influence of various plan design features and selected social norms on participation
and contribution choices have been covered. Decision-making biases and behavioral tendencies
have also been discussed along the way, when their mention facilitated an understanding of
observed behaviors. Behaviorists have identified other decision-making biases may also
influence participation and contribution; these are briefly discussed below.
Anchoring
Anchoring refers to the tendency to make small adjustments from a reference point, even though
large adjustments are more appropriate (Tversky & Kahneman, 1974). Research has shown that
even irrelevant reference points can have significant influence over choices (Ariely, Loewenstein
& Prelec, 2003).
Determining one’s optimal savings rate is a complex task requiring several economic
assumptions and is likely beyond the ability of even those workers interested enough to try. It is
therefore unsurprising that a large percentage of participants search for or find a savings-rate
“anchor” or reference point—whether it be the employer match threshold, the automatic
enrollment contribution rate or some other percentage that signals a contribution rate. In addition
to saving at the default rate (as discussed above), participants often select a contribution rate
based on the maximum employer match (the match threshold), a multiple of five (which Hewitt
[2002] calls the “round number” heuristic) or a prior (outdated) plan maximum.54 These
tendencies are observed in Choi et al. (2006) where the distribution of specific participant
contribution rates of one company is presented. The most popular rates observed are the match
threshold, 5 percent, 10 percent and 15 percent.
“Naive Reinforcement Learning”
Investing experience also has an effect on employees’ savings rates. Choi, Laibson, Madrian and
Metrick (2009) specifically find that one standard deviation increase in return for the year is
associated with a .13 percentage-point increase in savings rate and that an increase in return
volatility (again, by one standard deviation) predicts a .16 percentage-point reduction in
contribution rate at the end of the year, and a reduction of just over twice that much at the end of
54 Benartzi and Thaler (2007) study participant contribution behavior in one plan that increased the permissible maximum
contribution from 16 to 100 percent after the Economic Growth and Tax Relief Reconciliation Act of 2001. They find the
percentage of (new) employees deferring 16 percent declined by 75 percent. They suggest employees who may have anchored to
the 16 percent plan maximum switched to the “round number” heuristic, deciding to save 10 or 15 percent.
© 2014 Society of Actuaries, All Rights Reserved Behavioral Research Associates 44 | P a g e
the following year. The researchers call this the “naive reinforcement-learning heuristic” and
define it as an illogical emphasis on personal experience as a basis for decisions.
Priming
As previously discussed, a lack of self-control is implicated as one potential reason for low
savings rates. Evidence suggests promising interventions to address these self-control problems.
In one experiment, Nenkov, Inman and Hulland (2008) use priming (exposure to a deliberate
stimulus with the intention of having an effect on experimental subject’s response to another
stimulus) to increase subjects’ consideration (the researchers use the term elaboration) of
potential outcomes prior to making a decision about contributing to a 401(k) plan. Subjects were
randomly assigned to a priming condition in which they were asked to consider the positive and
negative potential outcomes of contributing or not contributing to the 401(k) plan. They find that
individuals who are inclined to think about future outcomes prior to making a decision contribute
more than those who are not so inclined. More importantly, subjects who did not exhibit the
natural tendency to think about the future but who were in the priming condition contributed
nearly 60 percent more than their nonprimed counterparts.
In another experiment, researchers used virtual reality technology to present subjects with aged
visual images of themselves and found that subjects in this condition allocated more than twice
the amount to a retirement fund than subjects who were only exposed to current visual images
(Hershfield, Goldstein, Sharpe, Fox, Yeykelis, Carstensen, Bailenson, 2011). It seems that
thinking about an older self helps individuals overcome a tendency to overweight the desires of
the current self.
Retirement Plan Investment Decisions
Within most defined contribution plans, participants are fully responsible for determining an
appropriate asset allocation and selecting specific investments. They fully shoulder the risk and
enjoy (or suffer) the consequences of their selections. Although investment advisory services
may be available to assist participants, these services tend to be underutilized in most plans.55
Within defined benefit plans, asset allocation and investment selection are responsibilities of
retirement plan sponsors, who often retain experts to guide them through these decisions.
Giving participants full responsibility for managing all of their retirement plan investments has
become more commonplace over the last 30 years. Papke (2003b) reports that in 1985, while 90
percent of participants in defined contribution plans directed the investment of their own
contributions, less than half did so for employer contributions made on their behalf. More recent
data from PSCA (2011) shows that now, 98 percent of respondent plans permit participant
direction of their own contributions and 92 percent permit investment direction of company
contributions.
55 In the plan studied by Agnew (2006b), she finds that although 15 percent enrolled in the advice service offered, only 3 percent
used the service more than once during the time period analyzed.
© 2014 Society of Actuaries, All Rights Reserved Behavioral Research Associates 45 | P a g e
In this section about workers’ investment choices, current asset allocation statistics, predictors of
investment choices and observed anomalies, including participant investment in employer stock,
are reported.
According to VanDerhei, Holden, Alonso and Bass (2012), at the end of 2011, approximately 61
percent of 401(k) plan assets were invested in equities: 39 percent in equity funds, 8 percent in
company stock and the remainder through investments in balanced funds. In addition,
approximately 12 percent was held in bond funds, 11 percent in guaranteed interest contracts and
stable value funds, 4 percent in money funds and 20 percent in balanced funds. VanDerhei et al.
(2012) report that asset allocation varies by age, the investment option choice set and salary. (See
Table 10 below.)
Table 10. Average asset allocation of 401(k) accounts, by participant age, salary and
investment options (percentage of account balances,a 2011)
Equity
Funds
Target-
Date
Fundsb
Nontarget-
Date
Balance
Funds
Bonds
Funds
Money
Funds
GICsc/
Stable-
Value
Funds
Company
Stock
Plans Without Company Stock, and GICsc and/or Other Stable-Value Funds
Age Group:
20s 35.8% 38.7% 7.4% 9.5% 3.5%
30s 48.1% 24.0% 6.3% 12.1% 4.4%
40s 51.0% 17.9% 6.2% 14.2% 5.2%
50s 44.9% 16.9% 6.7% 18.6% 7.0%
60s 38.5% 15.4% 6.6% 23.8% 9.4%
Salary:
$20,000–$40,000 40.4% 26.4% 7.1% 13.6% 6.3%
>$40,000–$60,000 42.5% 22.2% 7.8% 15.4% 7.1%
>$60,000–$80,000 45.5% 20.0% 7.4% 15.2% 6.4%
>$80,000–
$100,000 47.8% 18.4% 6.4% 16.1% 5.9%
>$100,000 49.2% 14.9% 6.6% 17.3% 5.6%
All 45.6% 18.0% 6.5% 17.8% 6.7%
Plans With GICsc and/or Other Stable-Value Funds
Age Group:
20s 35.1% 27.9% 14.2% 8.1% 1.4% 7.4%
30s 44.8% 20.0% 10.1% 7.9% 1.9% 9.4%
40s 47.5% 14.5% 8.6% 8.5% 2.2% 12.4%
50s 40.2% 12.8% 8.6% 10.4% 2.6% 19.2%
60s 32.7% 11.4% 8.2% 11.5% 3.3% 27.4%
Salary:
$20,000–$40,000 31.7% 18.1% 12.6% 9.6% 2.1% 20.1%
>$40,000–$60,000 34.8% 13.5% 14.2% 9.3% 2.4% 18.9%
>$60,000–$80,000 38.7% 10.6% 13.8% 9.6% 2.5% 17.9%
>$80,000–
$100,000
42.3% 9.4% 11.7% 10.4% 2.4% 17.0%
© 2014 Society of Actuaries, All Rights Reserved Behavioral Research Associates 46 | P a g e
Equity
Funds
Target-
Date
Fundsb
Nontarget-
Date
Balance
Funds
Bonds
Funds
Money
Funds
GICsc/
Stable-
Value
Funds
Company
Stock
>$100,000 44.6% 9.0% 10.3% 11.6% 2.3% 15.7%
All 41.1% 14.0% 8.9% 9.9% 2.6% 18.4%
Plans With Company Stock
Age Group:
20s 27.3% 41.7% 5.2% 6.7% 2.9% 12.0%
30s 38.4% 22.6% 4.4% 9.5% 4.4% 15.4%
40s 38.5% 17.2% 4.0% 11.0% 5.5% 18.9%
50s 30.3% 14.2% 4.6% 14.7% 8.9% 21.5%
60s 25.0% 12.6% 3.9% 18.4% 14.7% 20.5%
Salary:
$20,000–$40,000 25.7% 17.7% 3.9% 12.8% 15.6% 19.4%
>$40,000–$60,000 32.9% 12.8% 2.8% 12.8% 11.3% 22.5%
>$60,000–$80,000 31.3%
14.4% 4.4% 13.7% 8.4% 21.7%
>$80,000–
$100,000
33.9% 11.1% 5.5% 13.0% 7.5% 21.8%
>$100,000 33.7% 10.9% 4.8% 15.8% 5.8% 21.7%
All 32.4% 16.1% 4.2% 13.6% 8.7% 19.7%
Plans With Company Stock, GICsc and/or Other Stable-Value Funds
Age Group:
20s 30.4% 20.8% 15.8% 4.9% 1.4% 7.1% 16.0%
30s 41.5% 12.3% 9.7% 6.5% 1.7% 8.7% 15.4%
40s 42.5% 8.1% 7.9% 7.0% 1.8% 12.0% 16.3%
50s 34.3% 6.6% 7.4% 8.5% 2.2% 19.8% 16.7%
60s 27.7% 6.1% 7.1% 8.7% 2.4% 30.0% 14.4%
Salary:
$20,000–$40,000 29.9% 8.0% 11.2% 6.2% 1.5% 20.2% 20.9%
>$40,000–$60,000 32.0%
7.0% 11.4% 7.1% 1.9% 20.1% 18.4%
>$60,000–$80,000 33.1% 7.0% 10.0% 7.1% 1.8% 20.8% 17.5%
>$80,000–
$100,000
36.2% 6.1% 10.2% 7.9% 1.7% 18.4% 16.3%
>$100,000 39.6% 6.2% 8.0% 8.0% 1.4% 17.6% 13.9%
All 35.7% 7.6% 7.8% 7.9% 2.0% 19.0% 16.0%
Original source: Tabulations from EBRI/Investment Company Institute (ICI) Participant-Directed Retirement Plan
Data Collection Project a Minor investment options are not shown; therefore, row percentages will not add to 100 percent. Percentages are
dollar-weighted averages. b A target-date fund typically rebalances its portfolio to become less focused on growth and more focused on income
as it approaches and passes the target date of the fund, which is usually included in the fund’s name. c GICs are guaranteed investment contracts. d Salary information is available for a subset of participants in the EBRI/ICI database.
© 2014 Society of Actuaries, All Rights Reserved Behavioral Research Associates 47 | P a g e
Note: “Funds” include mutual funds, bank collective trusts, life insurance separate accounts and any pooled
investment product primarily invested in the security indicated.
Note. From “401(k) Plan Asset Allocation, Account Balances, and Loan Activity in 2011,” by J. VanDerhei, S.
Holden, L. Alonso, & S. Bass, 2012, EBRI Issue Brief 380. Employee Benefit Research Institute.
While aggregate statistics are interesting, particularly when compared to the fairly common
60/40 allocation often observed in defined benefit plans, we are more interested in the underlying
individual decisions that make up the aggregate statistics. As can be seen in Table 11 below,
wide variation in asset allocation is observed.
Table 11. Asset allocation distribution of 401(k) participant account balances to equities,a
by age, percentage of participants,b 2011
Percentage of Account Balances Invested in Equities
Age Group Zero 1–20% >20–40% >40–60% >60–80% >80–100%
20s 9.4% 1.5% 2.3% 5.3% 19.6% 61.9%
30s 8.8% 2.8% 3.7% 7.7% 20.4% 56.6%
40s 9.4% 4.0% 4.8% 9.2% 31.3% 41.3%
50s 11.4% 6.2% 7.0% 20.2% 30.5% 24.7%
60s 16.2% 8.3% 13.3% 25.1% 16.5% 20.6%
All 10.8% 4.5% 6.0% 12.8% 25.4% 40.6%
Original source: Tabulations from EBRI/ICI Participant-Directed Retirement Plan Data Collection Project a Equities include equity funds, company stock and the equity portion of balanced funds. “Funds” include mutual
funds, bank collective trusts, life insurance separate accounts and any pooled investment product primarily invested
in the security indicated. b Participants include the 23.4 million 401(k) plan participants in the year-end 2010 EBRI/ICI 401(k) database.
Note: Row percentages may not add to 100 percent because of rounding.
Note. From “401(k) Plan Asset Allocation, Account Balances, and Loan Activity in 2011,” by J. VanDerhei, S.
Holden, L. Alonso, & S. Bass, 2012, EBRI Issue Brief 380. Employee Benefit Research Institute.
The asset allocations of longer-tenured employees are most likely a result of participants’ initial
investment selections as altered by cumulative performance over time. In other words, they may
not necessarily reflect recent, active asset allocation decisions.56 Therefore, additional insight can
be gained by reviewing the asset allocation of recently hired employees. VanDerhei et al.’s
(2012) analysis shows that recently hired participants were much more likely to hold balanced
funds and in particular target-date funds, funds that offer a mix of investments in various asset
classes that become more conservative with the passage of time. Sixty-eight percent of recently
hired participants held balanced fund investments in 2011, compared to just 29 percent of recent
hires in 1998 (VanDerhei et al., 2012). About three quarters of these new balanced fund investors
held target-date funds, and over three-quarters of target-date fund investors held more than 90
percent of their account balance in these funds, perhaps owing to the increased use of target-date
funds as investment default in automatic enrollment plans. VanDerhei et al. (2012) find that
target-date fund usage varied with the investment menu available to the participant and age
56 As discussed below, a small percentage of participants make investment changes.
© 2014 Society of Actuaries, All Rights Reserved Behavioral Research Associates 48 | P a g e
(younger participants are more likely to use these funds and to a greater extent than older
participants).57
The individual characteristics associated with asset allocation decisions have been the subject of
a number of research studies that include analyses of behavioral data (Bajtelsmit & VanDerhei,
1997; Goodfellow & Schieber, 1997; Agnew, Balduzzi & Sunden, 2003)58 as well as survey data
(Sunden & Surette, 1998; Bajtelsmit & Bernasek, 2001; Dulebohn, 2002; Bieker, 2008).59
Generally, these studies find that allocations to risky assets are positively related to income and
negatively related to age.60 Results related to tenure are mixed. While Agnew, Balduzzi and
Sunden (2003) find that tenure is positively related to equity allocation, Bajtelsmit and
VanDerhei (1997) look at the effect of tenure on fixed-income allocations and report a positive
relationship up to a point, after which declines are observed.
Although Dulebohn (2002) finds no gender effects in his survey work, generally, studies report
findings that suggest women are more risk averse than men (judging by their allocations to
equities), but certain researchers are reluctant to make that conclusion since some data sets are
missing potentially important variables such as marital status (Bajtelsmit & VanDerhei, 1997;
Agnew, Balduzzi & Sunden, 2003; Goodfellow & Schieber, 1997; Bieker, 2008).61 Regression
model estimates in Sunden and Surette (1998) show the importance of considering marital status
in addition to gender. Their models estimate that single women and married men are less likely
(than single men) to choose “mostly stocks.” They further estimate that the choices of married
men and married women do not differ significantly, but that married women are more likely than
single women to choose “mostly bonds.”
Bieker (2008) finds further that college attendees and individuals with longer planning horizons
are more likely to hold equities and that the likelihood of holding stock decreases with wealth.
No effects from race, marital status for men, ownership of risky assets outside the plan, home
ownership or employer size are found.
Effects of the Investment Option Menu on Participant Choice
As further evidence of the impact of decision-making context, in this section, we report research
results demonstrating that participant investment choices are strongly influenced by the choice
set offered in the plan. More specific observations are highlighted below.
57 In addition, Mitchell, Mottola, Utkus and Yamaguchi (2009) show that among participants who are defaulted into target-date
funds and those in plans that previously offered static allocation funds, women and participants with lower account balances are
more likely to be target-fund investors. This work also analyzes the use of target-date funds as exclusive versus nonexclusive
holdings. (Target-date funds were designed to be used as an exclusive investment since they offer participants a preselected mix
of funds in various asset classes that becomes more conservative with the passage of time.) 58 Bajtelsmit and VanDerhei (1997) study a sample of 20,000 active management employees from one employer. The data are
from 1993. Goodfellow and Schieber (1997) analyze data for 36,000 participants in 24 plans, and Agnew, Balduzzi and Sunden
(2003) study data for 7,000 participants in one plan from April 1994 to August 1998. 59 Sunden and Surette (1998) analyze 1992 and 1995 SCF survey data; Bajtelsmit and Bernasek (2001) use 1994 HRS data for
their work that analyzes allocations of total wealth as opposed to plan wealth. Dulebohn (2002) collects survey data from 795
college and university employees in a Midwestern state. Bieker (2008) uses 1998 SCF survey data. 60 However, Bieker (2008) finds a positive relationship between age and equity allocation. 61 See also Yilmazer and Lyons (2010), who explore the effects of family decision making on portfolio choice. They find
portfolio choices of married men appear unaffected by characteristics of their wives, but portfolio choices of wives are affected
by their relative control and spousal age difference.
© 2014 Society of Actuaries, All Rights Reserved Behavioral Research Associates 49 | P a g e
Mixed evidence that participants evenly allocate their assets among those available when
the choice set is small is observed (Benartzi & Thaler, 2001; Agnew, 2002; Tang,
Mitchell, Mottola & Utkus, 2010; Huberman & Jiang, 2006).
Closely related, asset allocation is correlated with the proportion of equity and fixed-
income funds (Brown, Liang, & Weisbenner, 2007; Benartzi & Thaler, 2001).
Individuals allocate evenly to three or four funds, regardless of the number available
(Huberman & Jiang, 2006).
Large choice sets can be overwhelming, with some researchers suggesting that high-
knowledge participants are more confused by large choice sets than low-knowledge
participants (Agnew & Szykman, 2005) and others suggesting the opposite (Kida,
Moreno & Smith, 2010).
Larger choice sets are associated with higher allocations to stable value and fixed income
investments (Iyengar & Kamenica, 2010), but there is evidence that subjective
knowledge levels may play a role (Morrin, Broniarczyk, Inman & Broussard, 2008; Kida,
Moreno & Smith, 2010).
In experimental research with UCLA employees, Benartzi and Thaler (2001) observed a strong
behavioral tendency to split investment choices evenly over the number of choices offered,
allowing subjects’ asset allocation to be easily influenced by the composition of the investment
option line-up. The researchers call this the “1/n” heuristic. In one two-condition experiment,
employees in the first condition were asked to choose from four fixed-income and one equity
option (a menu similar to the actual choices offered to UCLA employees). Employees in the
second condition chose from a menu of four equity options and one fixed-income option (a menu
designed to be similar to the choices offered to TWA pilots). In the first condition, the average
allocation to equities was 43 percent compared to 68 percent in the four-equity condition. The
authors supplement their experimental findings with an analysis of actual plan allocations using a
large database of plans and also with a time-series analysis of one plan’s asset allocation during a
period of investment option changes. Results of both of these supplemental analyses confirmed
their experimental results.
Brown et al. (2007) study 11-K data for nearly 900 firms during the period from 1991 through
2000. Similar to Benartzi and Thaler (2001), they find strong menu effects. Specifically, their
work estimates that increasing the portion of equity options from one-third to one-half predicts
an increase in contribution allocations to equities by approximately 7.5 percentage points.
In contrast to Benartzi and Thaler (2001) and Brown et al. (2007), who analyze experimental
survey responses and aggregate plan data, Tang et al. (2010) and Agnew (2002) analyze
individual-level data and also report menu effects. (See Figure 8 below.) Agnew (2002) studies
1998 individual-level contribution allocation data for one plan that offered four investment
choices: a guaranteed income fund, an equity-income fund, an S&P 500 index fund and company
stock. Agnew (2002) finds that less than 4 percent follow the “1/n” heuristic and less than 5
percent follow the modified “1/n” heuristic (the modified “1/n” heuristics explains even
allocations to all other options, excluding a company-stock offering).62 Further, her regression
62 Benartzi and Thaler (2001) identify this behavior and suggest that employees view company stock as a separate asset.
Company stock is separately discussed below.
© 2014 Society of Actuaries, All Rights Reserved Behavioral Research Associates 50 | P a g e
model finds income and tenure are negatively related to the tendency to follow the modified
“1/n” heuristic.
Figure 8. Allocation of individual participant portfolios in 401(k) plans
AM references actively managed.
Note. From “The efficiency of sponsor and participant portfolio choices in 401(k) plans,” by N. Tang, O.S. Mitchell,
G.R. Mottola, & S.P. Utkus, 2010, Journal of Public Economics, 94, 1073–1085. ©2010 by Elsevier B.V. Reprinted
with permission.
Huberman and Jiang (2006) offer an alternative view of participants’ investment decision-
making.63 First, they find that despite the number of options offered to them, participants actually
invest in a relatively small number of funds—between three and four. Further, participants tend
to allocate their contributions evenly among their chosen funds. Finally, contrary to the other
researchers previously mentioned, they do not find strong menu effects. In other words,
participants’ asset allocations are only slightly affected by the portion of equity options in plans’
investment menus.
Even though the low number of funds used by participants would suggest little demand for large
investment menus, the menus have grown by about 50 percent since 1999 when an average of 12
options was offered (PSCA, 2011). Now, an average of 18 options is offered to participants
(PSCA, 2011). Based on their research using a data set of nearly 600,000 participant accounts in
638 plans, Iyengar and Kamenica (2010) find that more funds are associated with higher levels
of money market and bond fund holdings (and lower levels of equity holdings).64 More
63 The data set analyzed by Huberman and Jiang (2006) includes nearly 500,000 participant accounts in about 600 plans
recordkept by one recordkeeper. The number of options offered within these plans ranged from four to 59. 64 The data set only includes participants who had made an active investment choice.
© 2014 Society of Actuaries, All Rights Reserved Behavioral Research Associates 51 | P a g e
specifically, they estimate that for every additional 10 funds, equity allocations decline by 3.28
percentage points and bond fund allocations increase by 1.98 percentage points. Further, the
authors find that an increase of 10 funds increases the likelihood participants will allocate
nothing to equities by 2.87 percentage points. (Just under 11 percent of participants in the data
set allocate nothing to equities.) While the authors offer rational explanations for the observed
behavior, data limitations prevent the authors from precisely identifying why participants choose
simple, easy-to-understand options from large choice sets.
In experimental research, Agnew and Szykman (2005) find that increased choice sets affect
individuals differently, according to their financial knowledge. Individuals with a higher degree
of financial knowledge appear to be more overwhelmed by large choice sets than lower-
knowledge subjects who show a high level of overload even when choice sets are smaller. In
other words, for individuals with lower levels of financial literacy, there is no difference in their
degree of overload no matter how large the choice set.65 The researchers find an increase in the
number of similar options offered is associated with an increase in default selection, which may
be caused by a high degree of overload.
Morrin et al. (2008) report results of an experiment designed to test whether the asset allocation
effects of choice set size differed between low- and high-knowledge individuals. Subjects with
lower levels of subjective knowledge significantly allocated more to equities (60.2 versus 28.7
percent) when the choice set was large (21-fund choice set versus three-fund choice set).
Effects of Investment Performance on Investment Choice
Ample research sets forth evidence of performance chasing as an investment selection approach,
but much of it is outside the retirement plan domain. However, Benartzi and Thaler (2007) offer
two examples of real-world evidence of performance chasing. Figure 9 below separately shows
the equity allocations of all and new participants calculated from recordkeeping data from one
vendor. The first observation that can be made is the significant difference in the allocations of
the two groups, an observation also noted above in the more recent data from VanDerhei et al.
(2012). As discussed below, once participants make their selections, a significant percentage of
them never make any changes. The second is the way in which new participants responded to
market performance when making their investment choices.
65 This is in contrast to Kida, Moreno and Smith (2010), who find that experienced investors are not negatively impacted by
larger choice sets and may in fact be less likely to choose when the choice set is limited.
© 2014 Society of Actuaries, All Rights Reserved Behavioral Research Associates 52 | P a g e
Figure 9, Panel A. The equity allocation of new versus all plan participants
Panel A displays the percentage of new contributions allocated to equities by new versus all plan participants.
“New” participants are those entering the plan in a given year. The chart was constructed from data provided by
Vanguard.
Note. Figure 9, panels A. and B. are from “Heuristics and Biases in Retirement Savings Behavior,” by S. Benartzi
and R.Thaler, 2007, Journal of Economic Perspectives, 21(3), 81–104. ©2007 Benartzi and Thaler. Modified with
permission.
Similar decision making is noted in Figure 9, Panel B; only here participant allocation to a
technology fund is shown. Participant allocation to the fund appears to move in lockstep with the
fund’s share price.
Figure 9, Panel B. Percentage of new participants selecting the technology fund
58%61% 59%
69%74% 74%
71% 72% 74%70%
54%52% 53% 53%55% 57% 58%
59%63%
65% 67%64%
-30%
-10%
10%
30%
50%
70%
90%
1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002
Perc
en
t o
f N
ew
Co
ntr
ibu
tio
ns i
n E
qu
itie
s
New Participants All Participants
© 2014 Society of Actuaries, All Rights Reserved Behavioral Research Associates 53 | P a g e
Panel B reports the allocations of new participants at a large plan that offers a technology fund. The left axis
displays the percentage of new participants allocating some of their contributions to the technology, and the right
axis shows the fund’s share price. Data were provided by Hewitt Associates.
Even when the returns of different funds are incomparable because they cover different time
periods, individuals will tend to choose the fund with the highest since-inception return. In an
experiment conducted by Choi, Laibson and Madrian (2010), the researchers create a scenario
where a failure to minimize fund fees is difficult to rationalize. In this experiment, each subject
selected investments for her hypothetical $10,000 lump sum from a menu of four S&P 500 index
funds (whose only difference was cost).66 In the control condition, where subjects received fund
prospectuses, almost no one allocated their “money” to minimize fees. Even in the various
treatment conditions, where fees were made quite salient, 90 percent of staff and college student
subjects and 81 percent of MBA subjects failed to minimize fees. The researchers note that
subjects had a strong tendency to select the funds with the highest return since inception, which
is interesting since the funds each had different inception dates (and therefore, the returns since
inception covered different time periods).
Status Quo Bias and Default Acceptance
Once participants make their initial investment selections, a large portion never make any
adjustments throughout the course of their working careers despite the effects of uneven market
performance and likely changes in risk tolerance and investment menu. Samuelson and
Zeckhauser (1988) were among the first to document the status quo bias in retirement plan
investment choice. The authors refer to a 1986 TIAA-CREF study reporting that less than 30
percent had ever made a change to their investment choices. Twenty percent reported making
exactly one change and 8 percent reported making more than one change.
Similarly, Ameriks and Zeldes (2004) report that over the course of the 10-year period ended in
1996, 44 percent of participants made absolutely no change to either the allocation of their
current contributions or their accumulated contributions. Table 12 reports their findings.
66 In an effort to elicit closer-to-true decision-making, the researchers rewarded subjects based on performance of their selected
portfolio over a specified time period extending beyond the experimental session.
© 2014 Society of Actuaries, All Rights Reserved Behavioral Research Associates 54 | P a g e
Table 12. Changes in quarterly asset and flow allocations, 1987–96, sample of participants
with flows in all 40 quarters (n=4, 782)
Asset
Allocation
Changes
Flow Allocation Changes
Count 0 1 2 3–5 6–10 11+ Total
0 44.3% 15.3% 6.6% 5.3% 1.1% 0.2% 72.8%
1 1.9% 4.2% 3.2% 3.5% 0.8% 0.1% 13.7%
2 0.5% 0.8% 1.1% 1.9% 0.7% 0.0% 5.0%
3–5 0.4% 0.6% 0.6% 2.4% 1.0% 0.1% 5.1%
6–10 0.1% 0.2% 0.3% 0.4% 0.7% 0.2% 1.9%
11+ 0.0% 0.1% 0.2% 0.4% 0.4% 0.3% 1.5%
Total 47.1% 21.2% 12.0% 14.1% 4.7% 0.9% 100.0%
“Assets allocation changes” are the number of quarters in which assets are transferred from any deferred annuity
investment account to another via an “accumulation transfer” transaction, or in which assets are transferred from a
TPA to a deferred annuity account. “Flow allocation changes” also reflect changes in the allocation of contributions
to any of the investment accounts.
Note. From “How Do Household Portfolio Shares Vary With Age?” by J. Ameriks & S.P. Zeldes, 2004, Working
Paper. Reprinted with permission.
Other researchers report low levels of activity in retirement plan accounts as well. Agnew,
Balduzzi and Sunden (2003) find that 87 percent of participants in the plan they studied made no
portfolio changes during the four-year period ended August 1998. Mitchell, Mottola, Utkus, and
Yamaguchi (2006) study trading patterns of 1.2 million participants in 1,500 plans. During 2003
and 2004, 80 percent of participants made no adjustments to their portfolios.67
An interesting question is whether participants who have been defaulted into a retirement plan
are more likely to make investment changes since the default choice may not comport with their
preferences. Benartzi and Thaler’s (2002) finding that the majority of people (80 percent) who
expressed an interest in choosing their own investments ended up preferring a portfolio selected
by a managed account service would suggest that participants may not be any more likely to
change their investments even when they have been selected by others. However, as reported
above, even four years after automatic enrollment, Choi et al. (2006) found that between 29 and
48 percent of participants remain in the default fund. This suggests that participants who are
initially defaulted into an investment may in fact be more likely to change from it.
Employer Stock
In 2011, the aggregate allocation to employer stock in 401(k) plans was 16 or 19.7 percent,
depending on the other funds offered alongside it (see Table 9 above). This highlights another
67 The most commonly found attributes of those who do trade are: male, higher income, older, longer tenure (Agnew, Balduzzi
and Sunden, 2003; Mitchell et al., 2006). Mitchell et al. (2006) further finds traders tend to use the Internet, hold a greater
number of funds and invest in actively managed funds. The presence of employer stock in a plan is associated with increased
trading levels. Choi, Laibson and Metrick (2002) provide evidence that the introduction of web-enabled trading increases trading
levels, and in contrast to Agnew (2006a) and Mitchell et al. (2006), they find younger individuals are more likely to trade.
© 2014 Society of Actuaries, All Rights Reserved Behavioral Research Associates 55 | P a g e
difference between defined contribution and defined benefit plans. In defined benefit plans, no
more than 10 percent of plan assets may be invested in employer securities; no such limitation to
control risk applies to defined contribution plans.
While a majority of participants invest less than 20 percent of their retirement plan account
balances in employer stock, a significant portion of each age group invests more, as evidenced in
Table 13 below.68 Agnew (2006a) finds that males are more likely to overinvest in company
stock, and Utkus and Young (2012) also find a small negative effect from education (a college-
educated employee holds 1.2 percentage points less than an employee without a college
education).
Table 13. Asset allocation distribution of participant account balances to company stock in
401(k) plans with company stock, by participant age
Percent of Participants,a, b 2011
Percentage of Account Balance Invested in Company Stock
Age
Group 0
1–
10%
11–
20%
21–
30%
31–
40%
41–
50%
51–
60%
61–
70%
71–
80%
81–
90%
91–
100%
20s 65.8% 8.9% 5.0% 3.9% 3.2% 5.0% 2.3% 1.0% 0.7% 0.5% 3.8%
30s 52.5% 13.6% 9.1% 6.3% 4.5% 4.2% 2.5% 1.5% 1.0% 0.8% 4.1%
40s 48.2% 14.9% 9.6% 6.9% 4.8% 4.1% 2.7% 1.7% 1.3% 1.0% 4.8%
50s 45.9% 16.6% 9.8% 6.8% 4.7% 3.8% 2.7% 1.8% 1.4% 1.1% 5.5%
60s 49.4% 15.7% 8.7% 5.9% 4.0% 3.3% 2.3% 1.6% 1.3% 1.1% 6.8%
All 51.0% 14.3% 8.8% 6.2% 4.4% 4.0% 2.5% 1.6% 1.2% 0.9% 5.0%
Original source: Tabulations from EBRI/ICI Participant-Directed Retirement Plan Data Collection Project a The analysis includes the 9.1 million participants in plans with company stock at year-end 2010. b Row percentages may not add to 100 percent because of rounding.
Note. From “401(k) Plan Asset Allocation, Account Balances, and Loan Activity in 2011,” by J. VanDerhei, S.
Holden, L. Alonso, & S. Bass, 2012, EBRI Issue Brief 380. Employee Benefit Research Institute.
Employees’ investment in company stock is especially troubling to behaviorists for at least two
reasons. First, single-stock investment involves company-specific risk (idiosyncratic risk) that
can easily be diversified away. Meulbroek (2002) estimates that on average, employer stock is
worth only 58 cents to the dollar.69 Second, an employee’s human capital (future income stream)
is already invested in the employer.
Why do employees invest in their employer’s securities? In some cases, employer contributions
may be made in securities, with transferability restrictions. However, regulation prohibits these
restrictions once an employee reaches three years of service, so it is unlikely these restrictions
account for all, or even most, investment in employer stock. Researchers offer several other
potential explanations.
68 Utkus and Young (2012) find that within a subset of Vanguard’s recordkeeping data, only 10 percent of participants invest
more than 20 percent in employer securities. 69 Meulbroek’s 2002 analysis is as of December 31, 1998. Updated analyses are unavailable.
© 2014 Society of Actuaries, All Rights Reserved Behavioral Research Associates 56 | P a g e
First, Benartzi (2001) finds through survey work that participants perceive the employer’s
contribution of employer securities as implicit investment advice. Regression models in Utkus
and Young (2012) estimate that when employer matching contributions are made in employer
securities, participant allocations to company stock are 17 percentage points higher. The effect
from other employer contributions in company stock is an increase of 12 percentage points.70
Some of these increases likely result from transferability restrictions, but again, it is difficult to
believe restrictions account for all of the increases.
Second, Huberman (2001) posits that employees may ignore portfolio theory and invest in
employer stock because it is familiar to them (in the sense that they are familiar with their
employer). Huberman, Iyengar and Jiang (2007) note that lower-income employees (those
making below $40,000) are more likely to participate in plans that offer company stock, possibly
because at least one option in the plan is familiar to them.
Third, Benartzi and Thaler (2001) offer that employees mentally account for employer stock as a
separate asset class. They observe that when employer stock is available in plans, employees
naively allocate the remainder of their assets between equities and fixed income. In other words,
it appears that employees do not consider employer securities as a component of their equity
holdings but instead as a separate asset class.
Fourth, employees may be chasing performance. Using plan-level data from public company
filings, Benartzi (2001) is the first to study the relationship between past returns and employee
allocations to company stocks. He finds that past 10-year returns are positively related to current
contribution allocations to company stock. Benartzi’s work is further confirmed by Liang and
Weisbenner (2002) who study a larger number of companies over a longer time period (also
using plan-level data from public filings). Huberman and Sengmuller (2004) continue this work
and find that past three-year returns predict higher contribution allocations and transfers. They
also reveal that while employees react positively to favorable past returns, the converse is not
true: Unfavorable past returns do not seem to motivate a reduction in employer stock holdings.
Agnew (2002) and Choi, Laibson, Madrian, and Metrick (2004b) study individual-level data and
draw the same conclusion that employees’ investment decisions are driven in part by past
returns. Choi et al. (2004b) also find that favorable past returns predict employees’ reallocation
of employer stock holdings to other equities.
Fifth, evidence suggests that employees simply do not understand the risk associated with
company stock. Benartzi (2001) reports that just 18 percent of respondents to a John Hancock
Financial Services survey know company stock is riskier than a diversified stock fund. Mitchell
and Utkus (2003) also report Vanguard survey results. Vanguard Group (2002) finds that even
employees who understand individual stocks are riskier than a diversified stock fund don’t think
their company stock is riskier, suggesting some overconfidence.71
70 It is possible (and even likely) that inertia accounts for some company stock investments originating from employer
contributions. 71 The declines of Enron, WorldCom and Global Crossing, all of which occurred in 2002, could have illuminated the risks of
holding employer stock. However, Choi, Laibson and Madrian (2005) set forth a detailed analysis casting doubt on the effects of
this media-provided education on actual investment behavior. They estimate only a 2 percentage-point decline in employer-stock
holdings as a result of the publicized falls of Enron, WorldCom and Global Crossing.
© 2014 Society of Actuaries, All Rights Reserved Behavioral Research Associates 57 | P a g e
Leakage
For purposes of this section, leakage refers to employees’ preretirement removal of assets from
retirement-designated accounts in the form of a loan, hardship withdrawal or “cash-out.” While
loan and hardship withdrawal decisions only occur within the context of defined contribution
plans, participants covered by either type of retirement plan (defined benefit or defined
contribution) are often offered the option of taking a lump-sum payout prior to reaching normal
retirement age.72 In this section, we report on research that explores these three sources of
leakage which drain employees’ retirement resources.
Withdrawals
Over half of participants in defined benefit plans have access to lump-sum distributions upon
separation from their employer (U.S. Department of Labor, 2007). In addition to access to
retirement funds at job separation, defined contribution plan participants may also have access to
age-specific in-service withdrawals or “hardship” withdrawals.73 Over three-quarters of defined
contribution plans permit age-based in-service and financial hardship withdrawals (PSCA, 2011).
An estimated 14 to 16.2 million participants in the 2004 SIPP panel had received a withdrawal
(of any type) from a retirement plan (also of any type) through 2006 (Copeland, 2009, and
Purcell, 2009b).74 The median amount of all distributions received is relatively small at $10,000
(in 2006 dollars) and has shown a downward trend when amounts are aged according to when
they were received. A majority (54.5 percent) of the most recent distributions reported were
received by respondents who were 40 or younger at the time of the distribution (Copeland,
2009).
Participants who receive distributions prior to retirement generally face two alternatives: They
can roll the proceeds into a tax-deferred retirement account or they may use it for other purposes.
When the proceeds are not rolled into another tax-deferred retirement account (such as an
individual retirement account, or IRA), the withdrawal becomes a cash-out and participants must
pay a 10 percent penalty (in addition to taxes) if they are younger than 59 1/2.
As evidenced in more detail in Table 14 below, younger, nonwhite, unmarried, lower-income
and less-educated recipients were more likely to spend all or a part of a distribution (Purcell
2009b). Purcell (2009b) also finds that smaller distributions and those received prior to 1990
were more likely to have been partially or fully spent. Finally, distributions that were the result
of some type of involuntary event (such as sickness or employer closure) were more likely to be
spent.
72 Although loans are permitted in cash balance plans, they are rarely offered due to administration complexities. 73 For a good summary of the rules related to lump-sum distributions, hardship withdrawals and loans, see U.S. GAO (2009). 74 Although both researchers use 2004 SIPP data, Copeland (2009) includes respondents 21 and older who have left a job but not
retired. Purcell’s (2009b) 16.2 million includes all respondents 21 and older.
© 2014 Society of Actuaries, All Rights Reserved Behavioral Research Associates 58 | P a g e
Table 14. Disposition of most recently received lump-sum distribution, lump sums received
between 1980 and 2006 by individuals younger than 60
Received a
distribution
(thousands)
Rolled over
entire amount
(percent)
Saved some of
the distribution
(percent)
Spent entire
distribution
(percent)
Age when received
Under 35
35 to 44
45 to 54
55 to 59
6,003
4,005
2,859
1,047
38.0
49.1
50.1
57.9
45.0
37.9
41.3
34.3
17.0
13.0
8.6
7.8
Race
White
Other
12,219
1,695
47.1
31.2
40.0
51.5
12.9
17.3
Sex
Male
Female
6,532
7,382
46.7
43.8
40.4
42.3
12.9
13.9
Marital status in 2004
Married
Not married
8,851
5,063
50.3
36.2
38.1
47.2
11.6
16.6
Education
High school or less
Some college
College graduate
3,072
5,375
5,463
30.2
40.2
58.4
52.7
46.0
30.6
17.1
13.8
11.0
Monthly income in 2006
Lowest income quartile
Second-lowest quartile
Second-highest quartile
Highest income quartile
3,479
3,486
3,472
3,477
36.2
35.9
44.4
64.2
47.6
48.7
43.3
26.1
16.4
15.4
12.3
9.7
Amount of distribution
Less than $5,000
$5,000 to $9,999
$10,000 to $19,999
$20,000 or more
4,972
2,388
2,277
4,278
26.1
47.0
47.7
65.0
52.9
39.2
40.4
29.8
21.0
13.8
11.9
5.2
Reason for distribution
Retired or quit job
All other reasons
7,511
6,402
51.0
38.3
36.1
47.6
12.9
14.1
Year of distribution
1980 to 1989
1990 to 1999
2000 to 2006
1,794
4,846
7,274
37.4
46.9
45.9
44.4
40.8
41.1
18.2
12.3
13.0
Total 13,914 45.2 41.4 13.4
Source: CRS tabulations from the Survey of Income and Program Participation
Notes: Monthly income is person’s average income over four months in 2006. Quartile rank is based on income of
individuals who received a lump sum between 1980 and 2006 before age 60. Individuals with total monthly
individual income of less than $1,464 in 2006 were in the fourth (lowest) income quartile. Those with income of
more than $4,754 were in the first (highest) income quartile. Median total monthly individual income among those
who had received a lump sum was $2,876. Lump-sum distributions were adjusted to 2006 dollars.
© 2014 Society of Actuaries, All Rights Reserved Behavioral Research Associates 59 | P a g e
Note. From “Pension Issues: Lump-Sum Distributions and Retirement Income Security,” by P. Purcell, 2009b,
Library of Congress Congressional Research Service. Reprinted with permission.
Other research delves exclusively into withdrawals from defined contribution plans and
separately analyzes cash-outs and hardship withdrawals. While the individual consequences of
cash-outs and hardship withdrawals may be significant, U.S. GAO (2009) reports that less than
10 and 7 percent of 401(k) participants (a year) are affected by cash outs and hardship
withdrawals, respectively.75 The 2006 aggregate amounts involved are also relatively small:
Approximately 2.7 percent of 401(k) assets are withdrawn at job change and .3 percent due to
hardship. U.S. GAO (2009) reports a median cash-out of $4,166 and a median hardship
withdrawal of $3,123 in 2006.
Cash-outs from defined contribution plans tend to be taken by participants who are younger, with
lower incomes and lower account balances (Munnell, 2012, Aon Hewitt, 2011).76,77 The
consequences can be significant. The EBRI/ICI 401(k) Accumulation Projection Model
estimates that cash-outs at job change reduce the median estimated replacement rate for lower-
income participants in voluntary plans by 21 percent (Holden & VanDerhei, 2002).
In-Service Withdrawals
Plan recordkeepers report that between 6.9 percent (Aon Hewitt, 2011) and 4 percent (Vanguard
Group, 2013) of participants took in-service withdrawals (which include hardship withdrawals)
in 2010 and 2011, respectively.78 Aon Hewitt’s data show that in-service withdrawals have
trended upward from 2006, when 4.9 percent of participants took them. Approximately 20
percent of these withdrawals are hardship withdrawals. Vanguard Group (2013) and Fidelity
Investments (2010) report that approximately 2 percent of participants took hardship withdrawals
in 2012 and during the year ended June 30, 2010, respectively.79 Demographics related to
hardship withdrawals include:
Forty-five percent of prior-year recipients took another hardship withdrawal in the
current year (Fidelity Investments, 2010).
Participants between the ages of 35 and 55 are more likely to take hardship withdrawals
(Fidelity Investments, 2010).
Women earning between $20,000 and $40,000 were twice as likely to take a hardship
withdrawal as were men in the same income bracket (Fidelity Investments, 2010).
The average hardship withdrawal at Fidelity was $6,000 (measured over the year ended June 30,
2010), which is similar to that reported by Aon Hewitt (2011), which was $5,510 in 2010. The
75 U.S. GAO’s (2009) report is based on SIPP data from the 1996, 2001 and 2004 panels for 401(k) participants between the ages
of 15 and 60, inclusively. 76 For a more complete analysis of withdrawal recipients, see Butrica, Zedlewski and Issa (2010). This work uses recent SIPP
data to analyze the recipients of and reasons for preretirement cash-outs and withdrawals. 77 Munnell (2012) is based on the most recent SCF data. 78 Aon Hewitt’s data is derived from the behaviors of 1.8 million participants in 110 large defined contribution plans. Vanguard
reports on the activities of 3 million participants in 1,700 plans. 79 Fidelity’s analysis is based on the behaviors of 11 million participants in nearly 17,000 plans as of June 30, 2010.
© 2014 Society of Actuaries, All Rights Reserved Behavioral Research Associates 60 | P a g e
most common reasons for hardship withdrawals are: to prevent eviction (50.4 percent), education
costs (12.6 percent), medical costs (12.6 percent), past-due bills (7 percent), home purchase (6.3
percent) and tax payments (4 percent) (Aon Hewitt 2011).
The reasons for hardship are also often given for any withdrawal, as reported in Butrica,
Zedlewski and Issa (2010). Amromin and Smith (2003) posit that withdrawals with penalties are
perhaps rational attempts by liquidity-constrained households to smooth consumption as opposed
to “squandering of pension assets.”80
Loans
Loans can be another form of leakage, especially if there is a default, as is typically the case
when a terminating employee has an outstanding loan (Lu, Mitchell, & Utkus, 2010).81 U.S.
GAO (2009) estimates leakage attributable to plan loans in defined contribution plans to be $8
billion in 2006; however, some researchers suggest that future leakage could be much more
significant due to the recent economic downturn. Litan and Singer (2012) estimate that loan
leakage could increase to as much as $37 billion.
Citing the U.S. Department of Labor, VanDerhei et al. (2012) report that plan loans represent a
negligible portion of plan assets. This is despite the fact that most participants are in defined
contribution plans, which permit loans. In EBRI’s database, 87 percent of participants had access
to loans in 2011 (VanDerhei et al., 2012), but access varied widely by plan size. Thirty-four
percent of very small plans (10 or fewer participants) offer a loan provision, whereas 93 percent
of plans with 10,000 or more participants do. Approximately 20 percent of participants who have
access to plan loans take advantage of the provision; again, this varies by plan size, ranging from
19 to 24 percent (VanDerhei et al., 2012). At the end of 2011, the median and average loan
amounts outstanding were $3,785 and $7,027, respectively, which were slightly higher than prior
year amounts (VanDerhei et al., 2012).
Workers in the middle of their careers are more likely to borrow from their retirement-plan assets
(VanDerhei et al., 2012; Lu & Mitchell, 2010; Aon Hewitt, 2011). When demographic and other
variables are controlled for, tenure is positively related and compensation is negatively related to
having a loan (Beshears, Choi, Laibson, & Madrian, 2011; Utkus & Young, 2011). Loan
amounts, expressed as percentages of total balance, correlate positively with compensation and
show a tendency to be larger among middle-age workers (Beshears et al., 2011). Utkus and
Young (2011) also find that loan-taking is related to lower levels of financial literacy and
education, as well as the failure to fully pay credit card balances each month.82
The loan provisions also play a role in both the propensity to take a loan and the amount of the
loan taken.83 As expected, the higher the interest rate charged, the lower the likelihood of taking
80 Amromin and Smith’s (2003) work is based on information contained in 10 years of tax returns (1987 through 1996) for a
representative cross-section of 88,000 returns in 1987. 81 Their work is based on three years (July 2005 through June 2008) of recordkeeping data from Vanguard for 959 plans. Most of
the analysis relates to the behaviors of nearly 104,000 participants who severed employment with an outstanding loan. 82 Utkus and Young (2011) analyze 900 participant survey responses collected in August and September 2008. 83 For a thorough discussion of loan provisions, see Beshears et al., (2011, Section III). For most of their work discussed here,
they use data from Aon Hewitt.
© 2014 Society of Actuaries, All Rights Reserved Behavioral Research Associates 61 | P a g e
a loan. When interest rates are higher, the loan amounts tend to be higher (Beshears et al., 2011;
Lu, Mitchell, & Utkus, 2010). Other relationships include: loan duration (less than five years
relates to lower loan balances), number of loans permitted (probability of taking a loan is higher
in plans that permit two loans, as compared to plans that permit only one or more than two (a
finding dissimilar to Lu and Mitchell [2010], who find a positive relationship between the
number of loans permitted and the likelihood of having a loan), and permitted loan purpose (the
existence of loans to purchase a home are negatively related to loan utilization).
Additional insight is gained by considering the reasons people take loans. Using SCF data,
Beshears et al., (2011) find that the most common uses of loan proceeds are to purchase (or
improve) a home or other durables such as an automobile or appliance, and to pay for
educational, medical and occasional expenses (such as a wedding). Utkus and Young (2011) also
find that 40 percent of survey respondents used the proceeds for debt consolidation.
Borrowing from one’s retirement plan is rather innocuous, assuming the loan is repaid. The
EBRI/ICI 401(k) Accumulation Projection Model estimate a relatively minor reduction in
median replacement rates (less than half a percentage point) as a result of loan taking in
voluntary retirement plans (Holden & VanDerhei, 2002). Some researchers even suggest that
certain groups would be better off if they accessed this form of credit to a greater extent (see G.
Li & Smith, 2010). The risk is that the loan is not repaid (and becomes a withdrawal with taxes
and penalty). Lu, Mitchell and Utkus (2010) find that 80 percent of terminating employees with
an outstanding loan default. Employees with low levels of total wealth, income and retirement-
plan assets show a higher probability of defaulting (Lu, Mitchell & Utkus, 2010).
The Effects of Workplace Financial Education
Empirical evidence presented thus far points to an opportunity for improvement in retirement-
related decision-making, the importance of which is heightened in an environment where
employees are primarily responsible for their financial security in retirement. Previously,
employees’ lack of knowledge about the types and features of plans offered by their employers
has been covered. In this section, additional evidence related to workers’ financial literacy, the
implications and the results of educational attempts to improve it are presented. Readers will
note that low levels of financial literacy are undisputed and that all but one study documents the
expected relationship between financial literacy and financial decision-making. Positive effects
of workplace financial education are reported, but most of the research analyzing actual
behavioral change (rather than self-reported surveys) suggests the effects are statistically
insignificant or small in absolute terms.
Financial Literacy
Americans’ low level of financial literacy is well documented. For example, see Bernheim
(1995, 1998), Hogarth and Hilgert (2002), Moore (2003), Harris Interactive Inc. (2005), and
Lusardi, Mitchell and Curto (2010).84 Lusardi and Mitchell (2007a, 2007b, 2008, 2009, 2011)
84 For a more complete list of work related to financial literacy, see Huston (2010). This paper, as well as Hung, Parker and
Yoong (2009), discuss the measurement and definition of financial literacy.
© 2014 Society of Actuaries, All Rights Reserved Behavioral Research Associates 62 | P a g e
have made significant contributions highlighting the low level of literacy among older
Americans. Their development of three basic questions for inclusion in a supplement to the HRS
has demonstrated that:
One-third of older Americans could not correctly choose (from three alternatives: more
than $102, exactly $102 or less than $102) the correct amount in a savings account
earning 2 percent for five years.
Nearly a quarter could not determine whether they’d be able to buy more, the same or
less (than they currently could) at the end of one year if their money were in a savings
account earning 1 percent and inflation was 2 percent for that year.
Over half of respondents incorrectly thought that buying a single company stock provides
a safer return than a stock mutual fund (Lusardi & Mitchell, 2011).
Implications
This level of financial illiteracy is concerning, particularly given the transfer of retirement-
planning responsibility from employers to employees over the past three decades, if low levels of
literacy are associated with poor retirement preparation. Several researchers have empirically
demonstrated the existence of this connection. For example, see Bernheim (1998), who finds that
financial test scores are a significant predictor (along with college degree attainment) of
accumulated retirement wealth. Hilgert, Hogarth and Beverly (2003) also find statistically
significant relationships between saving and investment knowledge and saving and investment
behaviors in their work using University of Michigan’s Surveys of Consumers from November
and December 2001. Lusardi and Mitchell (2011) find their measure of financial literacy is
positively associated with the likelihood of planning for retirement, as do Hung, Parker and
Yoong (2009). Utkus and Young (2011) report that low levels of literacy are associated with a
greater propensity to borrow from one’s 401(k) plan. Brown, Kapteyn, Luttmer, and Mitchell
(2011) find that the ability to make informed annuity-related choices is also correlated with
higher literacy levels. However, Hung, Parker and Yoong (2009) find that financial literacy is
not associated with higher savings levels.
Effect of Education
In a 2011 survey of 458 employers, SHRM found that over 50 percent of employers offer
financial education to their employees; larger organizations (with between 2,500 and 24,999)
were twice as likely as organizations with less than 100 employees to offer it (72 vs. 36 percent)
(SHRM, 2012). Presumably, employers believe there are benefits derived from this financial
education.
Recent empirical evidence directly attributing observed behavioral changes to the effects of
workplace financial education programs is scant. Most research relies on self-reported data from
surveys. Main takeaways from this survey-based research are that workers report some benefit
from employer-based education (even workplace satisfaction, Hira & Loibl, 2005), and that the
effects are stronger in some demographic segments than others. Bernheim and Garrett (2003) and
© 2014 Society of Actuaries, All Rights Reserved Behavioral Research Associates 63 | P a g e
Lusardi (2004) find no effect from workplace financial education on total wealth. Employees
also have greater intention than execution (Clark, d'Ambrosio, McDermed, & Sawant, 2006).
One study suggests that employees have a relatively low level of retention (about 35 percent) one
year later (Clark, Morrill, & Allen, 2011). Bayer, Bernheim and Scholz (2009) find no effect
from printed materials and that seminar frequency is important. See Table 15 below for
additional information on survey-based research.
Table 15. Survey-based research on the effects of workplace financial education
Study Findings
Research based on individual surveys
Kratzer, Brunson, Garman, Kim, and Joo (1998)
178 employee post-seminar surveys collected in late
1997 or early 1998 (actual date not provided)
The researchers state, “Most of the workers report that
since their participation in the financial education
workshops, they make better financial decisions, have
increased confidence when making investment decisions,
changed their investment strategy by appropriately
diversifying or being more aggressive in their investment
choices, and have an improved financial situation.”
Also, they note that since a pre- and post-seminar survey
format was not employed, the results cannot be directly
attributable to the educational seminars.
McCarthy and Turner (2000)
855 surveys of employees covered by the Thrift Savings
Plan (collected in 1990)
The authors find “that written financial information
provided by employers increases the self-assessed
financial knowledge of employees and that individuals
who have a higher self-assessment of their financial
knowledge are more likely to contribute to their defined
contribution pension plan and more likely to invest in
risky assets.”
Muller (2001/2002)
640 respondents in the 1992 HRS wave
Retirement education is not associated with the overall
likelihood that participants will spend a retirement plan
distribution. However, it is associated with decreases in
the probability that participants age 40 and under will
spend a distribution and increases in the probability that
college graduates and women will.
Bernheim and Garrett (2003)
Cross-sectional telephone survey of 2,055 individuals
between the ages of 30 and 48, conducted in fall 1994
Participation rates are estimated to be 12.1 percentage
points higher when employment-based financial
education is offered.
The availability of financial education is associated with
higher savings rates, account balances and retirement
wealth at lower saving levels (below 50th percentile).
© 2014 Society of Actuaries, All Rights Reserved Behavioral Research Associates 64 | P a g e
Study Findings
No relationship between the availability of workplace
financial education and total wealth was found.
Muller (2003)
1,107 respondents in the 1992 HRS wave
When individuals are categorized according to their risk
aversion (low, moderate, high or extremely risk averse),
individuals who are highly risk averse are likely to
allocate 20 percentage points more to equities after
attending a retirement class.
Lusardi (2004)
1992 HRS survey data (nearly 5,300 observations,
approximately 500 indicating past retirement seminar
attendance)
Positive associations between retirement seminar
attendance and financial wealth and net worth are found,
particularly in the lower quartiles.
Maki (2004)
848 household survey responses, gathered in November
1995
“Respondents whose employers offered financial
education were 10 percent more likely to correctly
answer” (from a choice of stocks, bonds, savings
accounts or certificates of deposit) which one had
offered the best return over the past 20 years.
These respondents were also less likely to report not
knowing various features of their plans.
Hira and Loibl (2005)
700 surveys (collected in 1999) of employees of a
national insurance company who had attended a half-day
employer-sponsored seminar
A significant, positive relationship between seminar
participation and perceived improvement in financial
literacy was found. In turn, a significant, positive
relationship between subjective literacy improvement
and confidence and optimism about the future existed in
the results.
Clark et al. (2006)
633 pre- and post-seminar surveys and a subsequent
survey from a subset of this larger group of higher
education employees from a variety of schools who
attended a workplace retirement education seminar
between March 2001 and May 2002
A significant number of post-seminar survey respondents
indicated they intended to make a change in their
retirement goals, savings behavior and/or investments.
However, three months later, the authors find over a
majority of respondents who had planned to take action
had not done so.
Bayer, Bernheim and Scholz (2009)
1993 (n=910) and 1994 (n=861) employer telephone
surveys collected by KPMG for the KPMG Peat
Marwick Retirement Benefits Survey
Frequently conducted seminars are associated with
higher participation and contribution rates within the
nonhighly compensated employee group. No effect is
© 2014 Society of Actuaries, All Rights Reserved Behavioral Research Associates 65 | P a g e
Study Findings
found from the provision of printed educational
materials.
Clark et al. (2011)
Pre- and post-seminar surveys from attendees at five
companies collected between June 2008 and December
2009 (n=1,182), and follow-up surveys collected one
year later (n=187)
Their “analysis confirms that participants believe that
these [retirement] programs increased their financial
knowledge and in response to enhanced financial
literacy, many alter their retirement plans.”
One year later, respondents’ knowledge levels remain
higher than pre-seminar knowledge levels, but the
average retention rate is 34.8 percent of that recorded
immediately after the seminars.
In studies that directly measure the effect of employee education on attendees’ actual behaviors,
we see mixed results. Generally, the results are positive, and in some cases, small in absolute
terms. As above, employees’ intentions exceed their execution (Madrian & Shea, 2001b), and
interventions can have varying effects on different demographic segments. See Table 16 below
for additional information from these behavioral studies.
Table 16. Research using administrative data to assess the effects of workplace financial
education
Study Findings
Research that incorporates behavioral administrative data
Clark and Schieber (1998)
1994 participant and plan data from 19 firms ranging in
size from 700 to 10,000
Higher rates of participation are found when employees
are offered generic educational materials (15 percentage-
point increase) and tailored educational materials (21
percentage-point increase)
The availability of generic educational materials has no
impact on contribution levels, but tailored materials are
associated with a 2 percentage-point increase in
contribution rates.
Madrian and Shea (2001b)
Individual cross-sectional plan data from one company
(as of five different points from June 1999 through June
2000, n=29,011) and educational seminar attendance
data (indicating which employees attended a workplace
seminar, which was offered from January through June
2000, n=1,779)
The authors find that the seminars increase plan
participation and diversification into riskier assets
(stocks and bonds) from money market investments in a
statistically significant but small way. A majority of
attendees made no changes after the seminar.
Duflo and Saez (2003)
© 2014 Society of Actuaries, All Rights Reserved Behavioral Research Associates 66 | P a g e
Randomized experiment involving in excess of 6,000
nonfaculty staff at a university to assess the influence of
a number of factors, including the impact of benefit fair
attendance on plan enrollment
Researchers had access to administrative plan records
The authors find benefit fair attendance has a positive
effect on plan enrollment that is small in absolute terms.
They also find an upper bound of the effect of attending
the benefit fair on enrollment to be 14 percentage points
after 11 months.
Dolvin and Templeton (2006)
72 surveys of law firm employees, some of who attended
a workplace educational seminar in 2004, matched with
plan administrative data
Seminar attendees had better diversified and more
efficient portfolios with lower equity allocations than
respondents who had not attended a seminar.
Choi, Laibson and Madrian (2011)
Experimental survey (in 2004) of employees over the
age of 59 1/2 at one company (n=678), matched with
administrative records
Treatment condition included information related to the
cost of failing to take advantage of the (fully vested)
company-matching contributions
The difference between the contribution rate changes for
the control group and the treatment group (that received
the informational intervention) was statistically
insignificant.
Clark, Maki and Morrill (2014)
Surveys (collected in March and August 2011) and
administrative data for employees hired from 2008
through 2010 by one employer
A controlled experiment to test the effects of providing a
plan informational flier using nonparticipating subjects
(1,370 in each of three groups) provides the data for the
main findings reported here.
No statistically significant effect from the informational
treatment is found overall. However, subjects in the
treatment groups who were in the 18–24 age bracket
were twice as likely as similarly aged subjects in the
control group were to begin participating in the plan. A
similar, statistically significant positive effect for
subjects age 35–44 was also found. A statistically
significant negative effect was found for subjects older
than 45.
Financial Decisions at Retirement
Differences between the decision-making contexts of defined benefit and defined contribution
plans continue at the point of retirement with a general trend toward greater immediate
accessibility to one’s retirement wealth via the availability of a lump-sum payout option in both
types of plans, for which retirees generally exhibit a preference. However, life-cycle economic
models suggest better outcomes for most individuals if they were to annuitize some or most of
their retirement wealth. Researchers have explored a number of rational explanations for the
divide between the predictions of economic models and actual observed behaviors. Coming up
short, more recent efforts have focused on possible behavioral biases that may impact
annuitization decisions.
© 2014 Society of Actuaries, All Rights Reserved Behavioral Research Associates 67 | P a g e
In this section, we discuss the context in which the plan payout decision is made and report
research results from analyses of actual and planned payout behaviors at retirement. Next, we
reference research that explores possible rational explanations for the low levels of annuitization
before covering relevant work that seeks to reveal behavioral explanations.
Retirees covered by both types of plans (defined benefit and defined contribution) may face
similar distribution options, depending on plan provisions. Within the context of a defined
benefit plan, retiring participants are now offered more choices than ever. Many can choose an
annuity or a lump-sum payout, and in a relatively small percentage of plans, a combination of the
two may be selected. In 2010, nearly a quarter of participants covered by traditional defined
benefit plans in private industry could receive a lump-sum payout of their benefits, and virtually
all (96 percent) of nontraditional defined benefit plan participants could (U.S. Department of
Labor, 2011).85 However, in the past, it was much more common for defined benefit plans to
distribute benefits only in the form of an annuity. In 1989, only 2 percent of defined benefit plans
offered by medium and large private-industry firms permitted lump-sum distributions (U.S.
Department of Labor, 1990).
Retiring defined contribution plan participants may face more choices. Depending on plan
provisions, they may choose a lump-sum payment, regular installments, an annuity, deferment
(remaining in the plan) or a combination thereof. In a representative survey of workers retiring
between 2002 and 2007 who were covered by defined contribution plans, 70 percent reported
having a choice in the form of benefit distribution (Sabelhaus, Bogdan, & Holden, 2008). At the
same time that a lump-sum payout feature has become more prevalent in defined benefit plans,
the annuity form of payout from defined contribution plans has become less common. In 2010,
less than 17 percent of defined contribution plans offered an annuity as a form of distribution,
compared to nearly 38 percent of plans in 1998 (PSCA, 2011 and 1999).86
Distribution Choices within a Defined Benefit Context
The importance of the choice of payout options from a defined benefit plan cannot be
understated. Should an employee choose a lump-sum option, she takes on responsibility for (and
risks associated with) investing and withdrawal decisions during her retirement years—a time
when decision-making abilities of many, if not most, are on the decline (Agarwal, Driscoll,
Gabaix, & Laibson, 2009). Otherwise, she receives a fixed (or inflation-adjusted) monthly
payment for the rest of her life.
Given the stark difference in these two paths, one might expect more research in this area.
Although limited, the research presented here (in Table 17 below) shows wide dispersion in the
percentage of employees choosing a lump-sum distribution—from 12 to 96 percent—suggesting
significant contextual differences, which is highlighted in work by EBRI (Banerjee, 2013).
Within these contexts, correlation with individual characteristics is observed, but they do not
hold across all studies. As observed throughout this review, decision-making shortcomings are
evident. These include:
85 Seven percent of private-industry workers were in traditional defined benefit plans permitting a partial lump-sum payout with a
reduced annuity in 2010 (U.S. Department of Labor, 2011). 86 For comparison purposes, the U.S. Department of Labor (2010) reports 17 percent of participants in private-industry defined
contribution plans had access to an annuity form of payout in 2009.
© 2014 Society of Actuaries, All Rights Reserved Behavioral Research Associates 68 | P a g e
Choices appear to be influenced by the type of defined benefit plan (traditional vs. cash
balance) (Mottola & Utkus, 2007; Benartzi, Previtero & Thaler, 2011).
Employees exhibit “myopic extrapolation of stock market performance,” meaning that
they are more likely to choose an annuity after short-term unfavorable market
performance (Previtero, 2011).
Partial lump-sum distributions appeal to retirees when annuities are more valuable
(Chalmers & Reuter, 2012).
Table 17. Research on defined benefit plan choices at retirement
Study Findings
Banerjee (2013)
Banerjee analyzes 118,730 payout decisions between
2005 and 2010 in 84 defined benefit plans that included
both traditional and cash balance plans
Payout decisions are strongly influenced by plan design,
particularly related to the extent to which lump-sum
distributions are constrained. Reported annuitization
rates are:
98.8 percent in the case of no lump-sum distribution
restrictions,
94.5 percent in traditional plans with “strong” lump-
sum constraints,
44.3 percent in plans with no restrictions on lump-
sum payments, and
22.3 percent in cash balance plans with no
restrictions on lump-sum payments.
Annuitization rates increase with age and tenure.
Mottola and Utkus (2007)
Authors analyze distribution choices from two Fortune
500 plans from 2000 through 2006: 7,000 distributions
from a traditional defined benefit plan, and 21,000
distributions from a cash balance plan
Participants in the defined benefit plan were more likely
to choose an annuity, but most preferred the lump-sum
option. Twenty-seven and 17 percent of participants
choose the annuity form of payout in the traditional
defined benefit and cash balance plan, respectively.
Age is a strong predictor of an increased probability of
choosing an annuity. The authors estimate that a five-
year increase in age results in an 8 and 7 percentage-
point increase in the likelihood of choosing an annuity
form of distribution from the traditional benefit and cash
balance plan, respectively. Nearly half of the traditional
plan participants age 70 and older chose an annuity
compared with less than 20 percent for participants
between ages 55 and 60. An annuity was chosen by 62
percent of cash balance plan participants age 70 and
older.
Males and higher-income participants were less likely to
annuitize.
© 2014 Society of Actuaries, All Rights Reserved Behavioral Research Associates 69 | P a g e
Benartzi, Previtero and Thaler (2011)
Payout choices from 75 traditional defined benefit plans
and 37 cash balance plans
Similar to above, traditional defined benefit plan
participants were more likely to annuitize (53 percent)
than cash balance plan participants were (41 percent).
Regression estimates are that cash balance plan
participants are 17 percentage points less likely to
annuitize.
The authors suggest framing effects.
Previtero (2011)
Defined benefit plan payout choices using two data sets
are analyzed. The first includes payout data for the
period from 2002 through 2008 for over 103,000
retirees, enrolled in 112 different defined benefit plans
from 63 different companies. The second data set
includes 18,000 payout choices made by IBM retirees
between 2000 and 2009.
Forty-nine percent of participants in the first data set
chose an annuity.
Eighty-eight percent of retirees in the second data set
chose an annuity, and 6 percent chose a lump-sum
distribution. The remainder selected a combination of an
annuity and a lump-sum payout.
Previtero finds a strong negative relationship between
stock market returns and annuitization, estimating that an
increase of one standard deviation in stock market return
relates to a 6 percentage-point decrease in the probability
of annuitizing.
Chalmers and Reuter (2012)
Study of the payout decisions of over 32,000 retirees
from the Oregon Public Employees Retirement System
(PERS) between 1990 and June 2002 where retirees have
the unusual option of selecting a partial lump-sum
payment in exchange for a reduction in their lifetime
annuity
The annuity is calculated in two or three (depending on
hire date) ways, and annuitants are automatically paid
based on the highest calculated value. This feature
enables the researchers to analyze the impact of the
relative value of the annuity and compare its variation to
lump-sum payout preferences.
Eighty-five percent pass up the opportunity to take a
partial lump sum.
Retirees show greater preference for lump-sum
distributions precisely when the annuity is more
valuable, even controlling for market performance,
which could motivate a preference for lump-sum
payouts.
As standard economic theory might predict, the
following individuals were more likely to choose a
lump-sum payout: those with shorter post-retirement
lives, retirees who died within the first two years
following retirement, retirees who were less risk averse,
shorter-tenured PERS employees and lower-income
retirees.
Medill (2009)
1997 study of 1,607 payouts from the Nebraska Public
Employees Retirement System (a defined benefit
system)
Less than 4 percent chose an annuity at retirement
© 2014 Society of Actuaries, All Rights Reserved Behavioral Research Associates 70 | P a g e
Distribution Choices within a Defined Contribution Context
While some workers have access to installment payments and an annuity form of distribution,
most workers retiring from a company that sponsors a defined contribution plan have two
choices: take it or leave it (assuming the balance is over $5,000). In the research summarized in
Table 18 below, virtually all of which is based on survey work, a very strong preference for
lump-sum distributions is observed. While taking a lump-sum distribution may allow retirees to
take advantage of greater investment and tax planning flexibility, often, lower-cost investments
are available in a company-sponsored plan due to the plan’s purchasing power.
Table 18. Research on defined contribution plan choices at retirement
Study Findings
Utkus and Young (2010)
Post-termination behaviors of 133,300 plan participants
60 and older with average account balances ranging from
$110,000 to just under $150,000 (depending on year of
termination) who terminated between 2004 and 2008
In plans that permit partial distributions, a higher
percentage (27) of participants remained in the plan five
years after termination. In plans that did not permit
partial distributions, 18 percent of participants chose to
remain in the plan. However, only about 10 percent of
plans in their data set permitted partial distributions.
Table 19 summarizes their results.
Ameriks (2004)
TIAA-CREF participants taking distributions between
1978 and 2001
After introduction of other forms of distribution in 1989,
selection of the annuity form of distribution declined
steadily, and by 2001, it was chosen by less than half of
participants retiring from a TIAA-CREF defined
contribution plan.
Clark, Morrill and Allen (2014)
Survey of intentions of 620 older employees in two
companies that offer both a defined benefit plan and a
defined contribution plan (in 2008 and 2009)
Over 70 percent of respondents reported they would take
the annuity form of distribution from their defined
benefit plan and a lump-sum distribution from their
defined contribution plan.87
Sabelhaus, Bogdan and Holden (2008)
Two surveys of individuals who retired from companies
that sponsored defined contribution plans (between 1995
and 2000 in one survey [n=418] and between 2002 and
2007 in another [n=420])
Forty-seven and 54 percent of retirees with a distribution
choice reported taking a lump-sum distribution in the
2000 and 2007 surveys, respectively.
About a quarter of participants (with a distribution
choice) deferred distribution and nearly as many
87 The purpose of this study was in part to assess participant responses after attending an educational seminar, and these
percentages are post-seminar results. Pre-seminar results were slightly lower for the annuity form of payment from the defined
benefit plan (but still in excess of 70 percent) and approximately 7 percentage points higher for the lump-sum payment of defined
contribution assets.
© 2014 Society of Actuaries, All Rights Reserved Behavioral Research Associates 71 | P a g e
Study Findings
indicated they had purchased an annuity in both surveys.
Ten percent elected to receive installment payments (in
both surveys).
Of the respondents who reported they had received
lump-sum distributions, 62 percent said they had
reinvested all of the proceeds. Twenty-four percent said
they had reinvested some and spent some, and the
remaining 14 percent said they had spent it all.
Johnson, Burman and Kobes (2004)88
Data from the 1992 through 2000 HRS waves
Approximately one-third of participants leaving their
jobs after the age of 55 left their money in the plan;
another third rolled it into an IRA.
Approximately 15 percent withdrew or made other
(unknown) choices.
Four percent chose to annuitize their retirement assets.89
Individuals with lower account balances, lower income
and less schooling are more likely to withdraw their
assets. Table 20 summarizes these results.
Table 19. Participant distribution behavior as of year-end 2009, participants 60 and older
by termination year Year of Termination
2004 2005 2006 2007 2008
Percent of participants at year-end 2009
Rollover 57% 59% 58% 55% 47%
Cash 22% 20% 19% 19% 19%
Remain in plan 11% 13% 16% 19% 27%
Combination 6% 5% 5% 5% 5%
Installments 4% 3% 2% 2% 2%
Note. From “Distribution Decisions Among Retirement-Age Defined Contribution Plan Participants” by S.P. Utkus and J.A. Young, 2010. ©2010 by Vanguard. Modified with permission.
88 Hurd and Panis (2006) study this same data and find a vast majority (79 percent) of retirees keep their money in the company-
sponsored defined contribution plan. Another 14 percent cashes out, and 6 percent annuitizes. 89 The authors report that of those who left their jobs after the age of 65, 10 percent annuitized.
© 2014 Society of Actuaries, All Rights Reserved Behavioral Research Associates 72 | P a g e
Table 20. Disposition of DC plan assets for adults 55 and older, 1992–2000
Percent of
sample
Convert to
an annuity
Roll over
into IRA Withdraw
Leave to
accumulate Other
All 100% 4% 34% 15% 33% 14%
Gender
Male 53 4 36 12 33 15
Female 47 3 32 18 34 13
Age
55–59 54 2 34 13 31 19
60–64 34 4 35 16 37 9
65 and older 12 10 29 21 34 6
Education
Not high school
graduate 16 5 40 26 19 10
High school graduate 35 3 35 17 32 14
Some college 22 3 34 14 33 15
College graduate 26 5 29 8 42 17
Race
Non-Hispanic white 84 3 35 14 33 15
Non-Hispanic black 8 8 22 20 36 14
Hispanic 4 8 20 23 41 8
Other 3 0 50 26 19 12
Marital status
Married 74 4 34 13 34 15
Divorced or separated 8 5 33 11 34 17
Widowed 15 5 35 23 29 8
Never married 3 0 28 23 37 12
Children
No children 20 3 26 14 24 33
Any children 80 4 36 15 35 10
One child 9 4 49 14 20 13
Two children 24 1 41 14 35 10
Three or more
children 47 5 31 17 39 9
Employment status
Full time 29 1 32 14 38 15
Part time 12 4 40 12 33 11
Not working 57 5 34 17 31 14
Size of DC balance
Bottom quintile 20 1 24 38 23 15
Second quintile 20 2 35 21 33 9
© 2014 Society of Actuaries, All Rights Reserved Behavioral Research Associates 73 | P a g e
Percent of
sample
Convert to
an annuity
Roll over
into IRA Withdraw
Leave to
accumulate Other
Third quintile 20 7 36 6 38 13
Fourth quintile 20 7 35 5 40 14
Top quintile 20 2 40 5 33 21
Household net worth
Bottom quintile 20 4 23 33 23 17
Second quintile 20 5 28 16 36 14
Third quintile 20 3 33 11 35 19
Fourth quintile 20 3 41 8 43 7
Top quintile 20 5 45 6 31 13
Household income
Bottom quintile 20 5 36 27 21 12
Second quintile 20 3 37 18 34 10
Third quintile 20 4 31 13 37 15
Fourth quintile 20 4 30 9 40 18
Top quintile 20 4 37 8 37 14
Note. From “Annuitized Wealth at Older Ages: Evidence from the Health and Retirement Study,” by R.W. Johnson,
L.E. Burman, and D.I. Kobes, 2004, Final Report to the Employee Benefits Security Administration U.S.
Department of Labor. ©2004 by The Urban Institute. Reprinted with permission.
Single-Life or Joint-and-Survivor Annuity?
As has been seen throughout this paper, a significant portion of participants tends to passively
accept default choices. However, Johnson, Uccello and Goldwyn (2005) have a different finding
when it comes to the acceptance of joint-and-survivor annuities. In both defined benefit and
defined contribution plans, the default annuity choice is one that continues to pay benefits for as
long as one of the couple lives (a joint-and-survivor annuity). Should a married participant prefer
a single-life annuity, action (spousal consent) is required. Their study using HRS data analyzed
annuity choices found that nearly 30 percent of married men and 70 percent of married women
chose single-life annuities (Johnson, Uccello, & Goldwyn, 2005). They further review other
potential sources of survivor protection and when other sources are considered, only 7 percent of
married men and 3 percent of women choose single-life annuities.
Disposition of Lump-Sum Payouts
For insight into the disposition of lump-sum payouts, research by Moore and Muller (2002),
Hurd and Panis (2006), Purcell (2009b) and Verma and Lichtenstein (2006) is useful. Although
these works include analyses of distributions at any time in workers’ careers, one can reasonably
assume that a portion of activity of the 65-and-over population relates to retirees. Copeland
analyzes 2004 SIPP data to determine that approximately 44 percent of lump-sum distribution
recipients who were 65 or older contributed at least a portion to tax-qualified savings accounts
© 2014 Society of Actuaries, All Rights Reserved Behavioral Research Associates 74 | P a g e
(2009).90 Nearly 28 percent of recipients used some portion to pay for a home, business or debt
reduction. Another 23 percent reported that the funds were partially used for consumption
(Copeland, 2009).
The Annuity Puzzle
Most theoretical economic models predict that partial or full annuitization of retirement wealth is
rational for risk-averse individuals under a relatively wide range of conditions (Yaari, 1965;
Mitchell, Poterba, Warshawsky, & Brown, 1999; Davidoff, Brown, & Diamond, 2005; Gong &
Webb, 2008).91 And yet, a relatively low level of annuitization is observed. This difference
between theoretical predictions and observed behaviors has come to be known as the “Annuity
Puzzle.”
Early efforts to explain the annuity puzzle focused on rational explanations for individuals’
preferences.92 First, some retirees already receive sufficient annuity income from Social Security
and defined benefit plans (Dushi & Webb, 2004), but Brown, Casey and Mitchell (2008)
demonstrate this to be an insufficient explanation. Second, bequest motives may reduce the
optimal level of annuitization (Friedman & Warshawsky, 1990; Bernheim, 1991; Laitner &
Juster, 1996; Lockwood, 2012), but the importance of bequest motives within the American
population is not clear (Hurd, 1989; Brown, 2001; Kopczuk & Lupton, 2007). Third, a family
support system may enable risk sharing, thereby reducing the need for annuities (Kotlikoff &
Spivak, 1981; Brown & Poterba, 2000; Brown, 2001). Concerns about future large,
unpredictable expenditures such as health care may also reduce the attractiveness of illiquid
annuities (Sinclair & Smetters, 2004; Turra & Mitchell, 2004). Finally, the high loads due to
adverse selection could also explain the lack of annuity purchases, but researchers have found
this to be an incomplete explanation (Mitchell et al., 1999, Brown et al., 2008). In fact, it is
generally accepted that rational explanations fail to fully account for the low levels of
annuitization (Brown, 2007).
In the continued quest to solve the annuity puzzle, investigators have begun to explore
behavioral biases that may impact annuity-related decision-making. To date, researchers have
suggested several behavioral biases that may help solve the annuity puzzle: loss aversion (Hu &
Scott, 2007; Brown, 2007), an endowment effect stemming from loss aversion (Gazzale &
Walker, 2009), mental accounting (Hu & Scott, 2007; Brown, 2007) and framing (Agnew
Anderson, Gerlach & Szykman, 2008; Brown, Kling, Mullainathan, Wiens, & Wrobel, 2008;
Benartzi, Previtero and Thaler, 2011), among others (see Brown, 2007). Work by Brown et al.
(2011) also suggests that complexity and literacy may also have an impact on annuitization.
Below we briefly describe each of these behavioral biases and selected research papers.
Hu and Scott (2007) were one of the first research teams to set forth specific theories on
behavioral biases that may explain annuity purchase behaviors. In contrast to the standard utility
90 Verma and Lichtenstein (2006) report similar results from their analysis of 2003 SIPP data. 91 However, Reichling and Smetters (2013) find that when stochastic, rather than deterministic, survival probabilities are
modeled, annuitization is not appropriate for most households. 92 For a complete review of possible explanations, including those that are supply-related, see Brown (2007).
© 2014 Society of Actuaries, All Rights Reserved Behavioral Research Associates 75 | P a g e
model, they develop a behavioral model that reflects mental accounting (Thaler, 1985) and
cumulative prospect theory (Tversky & Kahneman, 1992), finding these biases theoretically
explain, at least in part, why more people don’t annuitize.93 As it relates to the annuity purchase
decision, mental accounting could be at work if individuals narrowly think about the annuity as a
separate and distinct gamble in which they can win only if they live long enough for the annuity
to pay off, without considering its potential effect on their overall lifelong consumption.
Cumulative prospect theory, as set forth by Tversky and Kahneman (1992), relates to several
behavioral biases and is based on the combination of three aspects of decision-making:
individuals’ use of a reference point against which a decision will be evaluated (rather than a
final outcome), the tendency to overweight extreme events with low probabilities of occurrence
and loss aversion. Its application to the annuitization decision is obvious. The reference point
would tend to be the status quo, which in the case of a single-premium life annuity, would be the
ownership of liquid assets equivalent to the annuity purchase price, the extreme event that could
be overweighted is an early death, and Hu and Scott (2007) simply explain the effect of loss
aversion. “Loss aversion always reduces the attractiveness of annuities. Simply put, an
actuarially fair immediate annuity will be rejected because the loss from possible early death
looms twice as large as a gain possible from living long enough to earn back the annuity
premium.”
How the annuity choice is framed matters (Agnew et al., 2008; Brown et al., 2008).94 Using a
controlled experiment with 945 nonstudent subjects95 in Williamsburg, Virginia, Agnew et al.
(2008) found that both women and men were less likely to choose an annuity when they had
viewed a five-minute slide show negatively framing annuities. However, only men were affected
by a negative investment frame. The authors suggested that perhaps women are only impacted by
negative frames that disconfirm prior beliefs. (The authors found that even after controlling for
risk aversion and literacy, women were more likely to choose annuities than men were.)96
Brown et al. (2008) also find strong framing effects in a between-subject online survey of 1,342
panel subjects, all over the age of 50. They frame four financial products (without using the
names of the financial products) using a consumption frame and an investment frame and ask
survey participants to indicate their preferences.97 They find a majority of individuals prefer the
annuity to the other financial products when the consumption frame was used. However, this was
not the case when the financial instruments were framed in investment terms. In this condition, a
majority of subjects preferred the alternative financial product.
The survey was structured to also test the effects of bequest motives, the loss of liquidity
(associated with the annuity option), the mortality premium and principal protection. The
researchers find preference for the annuity did decline (in both frames) when a strong bequest
93 Hu and Scott (2007) also posit other behavioral biases potentially affecting the annuity purchase decision but do not quantify
them. These include the availability heuristic, fear of illiquidity, hyperbolic discounting and risk vs. uncertainty. 94 Benartzi, Previtero and Thaler (2011) also suggest framing effects as an explanation of their finding that participants in cash
balance plans are less likely than participants in traditional defined benefit plans to annuitize. 95 Subjects ranged in age from 19 to 89 with a variety of income and education levels. The average ages of female and male
participants were 54 and 56, respectively. 96 The authors also tested default effects and found none but offered this may have been caused by their use of a weak default. 97 See Brown et al. (2007) for a complete description of the survey instrument.
© 2014 Society of Actuaries, All Rights Reserved Behavioral Research Associates 76 | P a g e
motive was present, but in most cases, a majority still preferred the annuity when the
consumption frame was used. The illiquidity of the annuity did not impact preference for the
annuity in the consumption frame condition, but this was not the case when the investment frame
was used. Additionally, the mortality premium had a positive effect when the consumption frame
was in force, but had little or a negative effect when the options were framed as investments. As
the researchers had hypothesized, principal protection is highly valued when the options are
framed as investments.
In the first round of surveys, no purchase price for the financial instrument was mentioned. This
was added to a follow-up survey; a summary of results is presented below in Table 21.
Table 21. Percent of respondents preferring annuities to alternative products
comparison of investment, consumption and modified consumption frames
Investment
Frame
%
(1)
Consumption
Frame
%
(2)
Modified Consumption
Frame
($100,000 initial
investment mentioned
for each product)
%
(3)
Life Annuity ($650 per month) compared to:
Traditional savings account
4% interest 21 72 68
20-year period annuity 48 77 79
$650 per month
35-year period annuity
$500 per month 40 76 73
Consol bond
$400 per month forever 27 71 70
N 321 352 406
Survey Arm IB IA IIA
Note. From, “Research Brief: Framing, Reference Points, and Preferences for Life Annuities,” by J.R. Brown, J.R.
Kling, S. Mullainathan, G.R. Wiens and M.V. Wrobel, 2008. ©2008 Brookings Institution. Reprinted with
permission.
Notes:
1. Each question described two fictitious men’s decisions for investing/spending in retirement and asked, “Who
has made the better choice?” All decisions were described in terms of amount and durations; the terms
“annuity,” “savings account” and “bond” were not used to label decisions.
2. The Investment frame (Arm IB) used terms such as “invest” and “earnings,” described periods in terms of
years, mentioned the value of the initial investment ($100,000 in every case), and alluded to the account value
at other points in the survey. The Consumption frame (Arm IA) used terms such as “spend” and “payment,”
described periods in terms of the individual’s age, and never alluded to an account or its value. The Modified
Consumption frame (Arm IIA) is the same as the Consumption frame, with the added mention of the initial
payment ($100,000 in every case) and added allusions to this account value at other points in the survey.
3. Standard errors range from 2.0 to 2.08 percentage points.
4. All survey respondents were 50 or older.
5. Survey Arms IA and IB were collected via an Internet survey in December 2007; Survey Arm IIA was collected
via a separate Internet survey in April 2008.
© 2014 Society of Actuaries, All Rights Reserved Behavioral Research Associates 77 | P a g e
Gazzale and Walker (2009) explore a possible endowment effect owing to the endowment of an
annuity in the case of the traditional defined benefit plan and a lump-sum payout in the case of
the cash balance plan. In their controlled laboratory experiment, 373 George Mason University
students played a game to earn either annuity income or stock wealth. In a third condition,
subjects earned nothing (no endowment). They were further allocated to either a sequential or
simultaneous framing condition to determine the end of the annuity payout stream. In the
sequential frame, subjects had to survive each round to continue, whereas in the simultaneous
frame, a single draw determined the length of participation. The authors suggest that the stock-
wealth endowment, sequential condition mimics the current frame for 401(k) participants.
The authors find participants in the annuity and no endowment treatment conditions were more
likely to choose the annuity form of payout than participants in the lump-sum condition,
regardless of the survival probability condition (sequential or simultaneous). They attribute this
to an endowment effect stemming from loss aversion (prospect theory). Overly simplified, the
endowment effect relates to the difference between the amount one is willing to pay for
something and the amount he is willing to accept for something (his endowment) (Thaler, 1980).
Research has found several circumstances where people have required a selling price that is
about twice the price they would pay for it (Kahneman, Knetsch, & Thaler, 1990; Carmon &
Ariely, 2000).
In addition, subjects in the sequential treatment condition were less likely to choose the annuity
option than subjects in simultaneous condition. When risk-related measures were included in
their regression model, the model estimates an effect of the sequential frame of lowering the
probability of choosing an annuity by 15 percentage points. They attribute this to subjects
overweighting the probability of dying early, relative to the probability of a long retirement
(Gazzale & Walker, 2009).
Recent experimental survey work by Brown et al. (2011) highlights the role of complexity and
literacy on consumers’ ability to value annuities. Specifically, over 2,000 subjects from RAND’s
American Life Panel traded off between a hypothetical Social Security annuity and a lump-sum
payment. They find that subjects valued the Social Security annuity more highly when asked to
give them up than they did when given the option to buy them. Further, a series of trade-off
inquiries produced uninformed, inconsistent choices among those with lower levels of literacy.
The team suggests additional research to further explore their hypothesis that the complexity of
the annuity decision may be a factor in consumers’ annuity purchase decisions.
Conclusion
It certainly is easy to frame a review of research that provides insights into employees’
retirement-related financial decision-making, knowledge and perceptions as a proverbial “glass
half empty.” As the retirement system in the United States has evolved over the last 30 years into
one in which individuals directly shoulder the responsibility for funding their retirement years,
we have had the opportunity to observe myriad shortcomings in their decision-making, as
covered herein. Workers value retirement benefits, but many don’t understand them. Albeit a
© 2014 Society of Actuaries, All Rights Reserved Behavioral Research Associates 78 | P a g e
small percentage, some don’t even know they are contributing to a retirement plan. While there
is certainly significant evidence that employees simply follow the “path of least resistance”
(Choi et al., 2002), which may mean they never begin saving if they happen to work for a
sponsor with traditional enrollment processes, there is also ample evidence that sometimes they
actively choose. Many times, choices, whether passively or actively made, are suboptimal with
costly consequences. Overwhelmed by choice, employees may fail to diversify and choose
investments that are too conservative. Or, they may simply split contributions among a
manageable number of options. They may ignore the fine print, look in the rear view mirror, and
invest in what has historically been the best performing fund, without regard for the risk
involved. Employer stock seems safe, possibly because it is familiar. Retirement savings are
often lost at job change with the potential effect of slashing retirement income by two-thirds.
None of this should be surprising because financial literacy is abysmally low and financial
education is often ineffective. Over 40 percent of baby boomers are at risk of not having enough
money in retirement (VanDerhei, 2012). Finally, few retirees purchase an annuity to protect
themselves in the event they outlast their money.
However, there is another frame. The retirement saving glass can also be viewed as one that is
half full with many stakeholders working hard to fill it. It has been only 30 years that workers
have been saving for retirement in 401(k) plans, the most popular type of defined contribution
plan. How would plan sponsor decision-making have been evaluated in the early 1900s—just 30
years after the first known defined benefit plans began? A frequent precursor to improvement is
identifying the nature and source of the problem. As evidenced by much of the research
described herein, researchers and industry leaders have made significant contributions doing just
that. They have observed the predictable, irrational (and human) tendencies that undermine fully
rational decision-making. But not only have researchers and industry leaders uncovered decision-
making shortcomings and quantified their potential impact, they have also shown ways to
motivate better choices and outcomes by exploiting the very behavioral tendencies that can
hinder retirement outcomes. Automatic enrollment and automatic salary-deferral increase
programs turn inertia into positive outcomes by increasing participation and contribution rates.
Automatic and simplified enrollment reduce choice overload, making it easier for employees to
become participants. Researchers are uncovering behavioral aspects of the annuity purchase
decision, showing that individuals’ perceptions are greatly influenced by framing, the
endowment effect and product complexity. Additional decision-making “prescriptions” and
improved outcomes are likely to follow.